Internal discussion memo · 7 November 2026
Internsive:
sell trust, not reach.
Internship platforms across India and the world, what stands between our demo and production, crafty moat ideas, and the selling point.
CALL: DO NOT BUILD YET. TEST FIRST.
Nothing here commits anyone to building. General research, not legal advice. Figures carry evidence labels; model outputs are marked as such.
Internsive · Working draftExecutive summary
The call: don’t fund a big build. Spend 2–3 weeks on evidence, then pilot one city by hand.
- Not close to production. A convincing prototype with no backend. Every company, reply time and applicant is sample data, and several on-screen claims are untrue today.
- The market is small and hasn’t paid anyone yet. Internshala and Unstop made about ₹77 Cr combined in FY25 and lost about ₹24 Cr. The leader reportedly changed hands at about ₹100 Cr. This is a low-cost, one-city business with options attached, not a national venture bet.
- The opening is real but unproven. The loudest student complaints are fake listings, silence and being sold to. Incumbents earn from upsells and from selling application volume, so they have reasons not to fix all three. Nobody has tested the alternative in India.
- The moat is a record, not an interface. Free capped signals are the best-evidenced cheap mechanic. A verified ledger of who replied, paid and finished is what compounds. A free college logbook is the distribution. Swipe mode is none of these.
- Your resume builder: build it free and small, after the core. Internshala’s is already free, so it is table stakes, not a moat.
The selling point
Student: “Checked company, paid stipend, always an answer, never a fee.” Company: “Fewer applications, from students who chose you, at one flat published price.”₹80–180 CrMonetised early-career pool today. Our model; no published estimate fits.
₹300–700 CrPlausible 3–5 year ceiling. Also modelled.
≈2.3×Revenue multiple in the Internshala deal. Valuation reported, unconfirmed.
≈4×Kanzhun EV/sales at 44% margin, the cleanest marketplace benchmark.
100–115Engineer-days to a legally defensible launch, about 5 months for one engineer.
₹23,300Monthly running cost at 10,000 sign-ups. Excludes salaries, lawyer and CA.
390–1,300Paying employers to break even at founder scale (₹50 lakh a year). Assumption.
0Moat mechanisms tested in an Indian internship market. This is the key risk.
Detail and sources on the following pages. ₹1 Cr = ₹10 million; 1 lakh = 100,000. USD at ₹90/$ (assumption).
Internsive · Discussion memo
How to read this
Every number carries its pedigree. Treat model outputs and search snippets as leads, not facts.
Evidence labels
| HARD | Government data, statute, regulator text or filing read directly |
| OBSERVED | Live public page read logged out on 7 Nov 2026 |
| COMPANY | A company’s own claim, page or survey, unaudited |
| PRESS | Press or data aggregator; underlying filing not seen |
| SECONDARY | Blog, roundup, personal notes or a competitor’s description |
| SNIPPET | Search-result summary only; page not opened. Check before quoting |
| REVIEW | One unverified user review. An anecdote, never a trend |
| STUDY | Academic or working paper |
| DERIVED | Our arithmetic on sourced numbers |
| ASSUMPTION | Judgement with no source |
Fix-and-add tags
FIXwrong or fake today ADDmissing LEGALrequired now LEGAL-2027required from May 2027, build at launch BLOCKERdo not launch without NICEnice to have
Legal tags are research readings. A lawyer must confirm them.
Contents
- Thesis and variant view
- Market size and break-even
- Comparables
- Competitor map: India, government, world
- Where incumbents are conflicted
- User pain
- The selling point
- Production gap: fix and add, from sign-up onward
- Resume builder verdict
- Khurafati moat ideas, ranked
- Ideas to kill
- Catalysts
- Risks and what would prove us wrong
- Staged plan with kill criteria
- Decisions needed from you
- Evidence gaps and source conflicts
Coverage
Six research tracks: India, global, user pain, production and law, moat mechanics, market economics. 232 cited links. The full memo holds every source and 20 logged conflicts between sources.
Biggest holes: Reddit was unreadable, competitors were only seen logged out, and nobody has measured how common scams are.
Thesis
Long verified outcomes, short registrations.
LONG
- Verified outcomes: who replied, who paid the stipend, who finished.
- Scarce intent: students choosing a few employers rather than spraying hundreds.
- Dense local supply: one city and one field before any breadth.
SHORT
- Registered-user counts, listing counts and thin college pages.
- Swiping as an identity, AI auto-apply, course upsells.
Why now
- The leader just changed hands. upGrad, an edtech group, took 90% of Internshala in Feb 2026 PRESS. That could distract it or push upselling, or simply better resource it. A hypothesis.
- Fresher hiring intent is rising: 70% in H2 2025 to 75% in H2 2026 COMPANY. TeamLease sells in this market, and it tracks any-fresher intent, not intern headcount.
- Employers report overload: 74% of Indian recruiters struggle to find qualified candidates, and applicants per job have “more than doubled” since 2022 COMPANY. LinkedIn sells hiring AI, so this is directional.
Why not now
- The modelled pool can’t support a large raise. Every comparable with a ₹45–194 Cr cost base lost money in FY25 DERIVED.
- Undergraduate enrolment fell 0.27% in 2023-24, described as the first decline since AISHE reporting began in 2011, mainly in arts, science and commerce HARD.
- A subsidised PM Internship Scheme rollout could reprice stipends.
- AI could cut intern seats in exposed roles. Which way it nets out is unknown.
Variant view
The market prices reach. The evidence says reach, on its own, is worth very little.
| Market believes | We believe | Evidence |
|---|---|---|
| Registered users are the asset | Worth little without a payer | Internshala: 34M+ registered COMPANY, ₹13 revenue each DERIVED, about ₹100 Cr valuation PRESS. Internships.com sold at about $5.50 per student in 2014 SNIPPET |
| More applications = better product | Volume is now the defect | 244 applications per job in 2025 vs 115 in 2022, Greenhouse, US-centric SNIPPET. A soft applicant cap cut applicants 11% with no discernible effect on hiring STUDY |
| Students are the second revenue line | That’s the conflict, and the biggest player gave it up | Job seekers were about 0.6% of Boss Zhipin revenue in Q2 2026, and shrinking HARD |
| AI matching is the next moat | AI is a commodity on both sides and adds noise. Proof gets scarcer | LinkedIn about 11,000 applications a minute, +45% in a year SNIPPET. Counterpoint: Naukri’s AI buyers paid 10–15% more PRESS |
| Mutual match or swipe is a fresh interface | No evidence of recruiters adopting and paying at scale | Sorce dropped the employer side altogether, Oct 2025 PRESS |
| SEO is the cheap channel | Eroding, even for strong owners | Shiksha billings fell 13% to ₹45.1 Cr in Q4 FY26 as Google referrals fell PRESS |
| Handshake proves a campus network is worth billions | Recruiting there is about $150M. The value came from redeploying a verified network | About $1.15B gross, about $1B of it AI training, Apr 2026 PRESS. One US case, PhD-heavy supply |
| Government schemes failed, ignore them | Best public funnel data, and a live competitor | Round 3 at ₹9,000 passed 30,000 joiners by Oct 2026 PRESS |
Where the market could be right
- Distribution may simply win. Naukri has 75%+ traffic share among peers and could give Campus away PRESS.
- Store ratings are high (4.0–4.7 on Google Play), so most users may not feel what low-star reviewers describe.
- Employer payment alone hasn’t yet built a large independent internship business in India. Dropping courses leaves monetisation unsolved.
Sources: MediaNama · Horton et al. · Kanzhun 6-K · Techpoint · MediaNama on Shiksha · Dealroom · ThePrint
Internsive · Discussion memo
Market size · Step 1: students
A hard base of 92 lakh graduates a year, but only about 14 lakh paid internships in our base case.
| Step | Low | Base | High | Label |
|---|---|---|---|---|
| Regular-mode graduates a year (UG, PG, diploma) | 92.4 lakh | 92.4 lakh | 92.4 lakh | DERIVED |
| Internship-relevant pool: × 2 final years | 185 lakh | 185 lakh | 185 lakh | ASSUMPTION |
| Share seeking a formal internship in a year | 16% | 25% | 40% | ASSUMPTION |
| Active seekers | 30 lakh | 46 lakh | 74 lakh | DERIVED |
| Share who get one | 30% | 50% | 60% | ASSUMPTION |
| Internships obtained, incl. unpaid and college-arranged | 9 lakh | 23 lakh | 44 lakh | DERIVED |
| Share paid | 40% | 60% | 75% | ASSUMPTION |
| Paid internships | 3.6 lakh | 13.9 lakh | 33.2 lakh | DERIVED |
Base: AISHE 2023-24, released 8 July 2026. 4.50 Cr enrolled, 1.10 Cr passing out HARD. About 2.5 years old in reference terms.
Three checks on this funnel
- Observed floor: Internshala’s about 30 lakh active applicants a year, which also includes fresher-job seekers COMPANY.
- Stated demand is far higher: 92.8% of students “seek internships” in the India Skills Report 2026 SNIPPET. Wanting one is not actively seeking one.
- Measured neighbour: NAPS 9.85 lakh + NATS 5.23 lakh = 15.08 lakh apprentices in FY25 PRESS. Close to our base case, and actually counted.
What this means
- No credible national count of internship seekers exists. Every total on this page is ours and should not be quoted as a market estimate.
- Paying for student acquisition is effectively ruled out: Internshala earns about ₹147 per active applicant DERIVED, so a sustainable CAC is perhaps ₹30–50 ASSUMPTION.
Market size · Step 2–3: money
The monetised early-career pool is about ₹80–180 Cr a year. Modelled, not measured.
Early-career revenue pool, three methods (₹ Cr a year)
Bar = low to high. Tick = base case, where the method has one. Scale runs to ₹750 Cr.
Top-down share
₹290–710
Employers × spend
₹76–417
Paid internships × fee
₹7–398
Our read today
₹80–180
3–5 yr ceiling
₹300–700
Reality check
- Internshala + Unstop = ₹77 Cr in FY25, including fresher jobs, competitions and training PRESS.
- upGrad’s own ambition: about ₹100 Cr in about two years, starting with 34M users PRESS.
All online recruitment
- Identified floor ₹2,587 Cr, mixing years: Info Edge ₹2,374 Cr (FY26), WorkIndia ₹78.7 Cr, Apna recruitment ₹57.1 Cr, Internshala ₹44.1 Cr, Unstop ₹33.1 Cr (FY25) DERIVED from PRESS.
- If Naukri holds 50–65% of revenue ASSUMPTION, the total is ₹3,650–4,750 Cr, about $405–530M DERIVED.
- LinkedIn India, Indeed India and foundit revenues weren’t found. They are the main uncertainty.
Published estimates don’t fit
- IMARC’s $80.0M (2024) covers recruitment software only.
- Ken Research’s $25.9B (2025) includes staffing payroll.
- “Over $1B” is an informal 2018 guess by Info Edge’s then CEO, who said he had no solid data.
Far option, not a base case: stipend flow-through
13.9 lakh × 2.5 months × ₹8,000 to ₹21,500 = ₹2,780–7,470 Cr. A 5% fee gives ₹139–374 Cr DERIVED / ASSUMPTION. No Indian precedent found, and holding stipends runs into RBI payment-aggregator rules.
Sources: Inc42 Internshala · Inc42 Unstop · Inc42 on upGrad · IMARC · Ken Research · MediaNama 2018. Method inputs are assumptions; see full memo.
Internsive · Discussion memo
Unit economics
At founder scale this pays with 390–1,300 employers. One city alone won’t do it.
| Case | Annual cost | Spend per paying employer | Paying employers needed | Revenue needed |
|---|---|---|---|---|
| Founder scale, list-price listings (1.5 × ₹2,999) | ₹50 lakh | ₹4,500 | about 1,300 | about ₹59 lakh |
| Founder scale, subscription pricing | ₹50 lakh | ₹15,000 | about 390 | about ₹59 lakh |
| Lean company, about 20 staff | ₹4 Cr | ₹15,000 | 3,137 | ₹4.7 Cr |
| Lean company, higher price | ₹4 Cr | ₹40,000 | 1,177 | ₹4.7 Cr |
| National, Internshala-sized costs | ₹50 Cr | ₹30,000 | 19,608 | ₹58.8 Cr |
Formula: cost ÷ (spend per paying employer × 85% contribution margin). All inputs ASSUMPTION, results DERIVED. Founder-scale rows are ours; the research model assumed 20 staff.
₹36 lakhA one-city beachhead: 800 employers × ₹4,500, even if every one paid.
₹980What Internshala earns per employer ever acquired. Its paying count is unknown.
₹10–20 CrValue of ₹5 Cr revenue at 2–4×. A real business, not a venture outcome.
How to size this bet
- Spend per employer is the most fragile input. Small employers already get free posts on several boards.
- ₹5 Cr of revenue would be about 3–6% of today’s pool. Reachable, but it needs hundreds of repeat payers.
- A ₹1,000 Cr outcome at 4× needs ₹250 Cr revenue, about three times today’s whole pure-play revenue.
- Small can work. Instahyre made ₹18.9 Cr revenue and ₹3.1 Cr profit in FY25; Cutshort ₹2.7 Cr with ₹9.8 lakh profit PRESS. Both are tech-only.
- Base rate on exits: internship-only boards tend to be absorbed cheaply. Praktikum.info sold for six figures in 2011 SNIPPET; Looksharp went to WayUp in 2017, terms undisclosed.
Implication: a low-cost business with options attached. Decide the end game openly (Decision 13).
Sources: Inc42 Instahyre · Cutshort pricing · Internshala blog, ₹2,999 · Deutsche Startups · PR Newswire
Internsive · Discussion memo
Comparables
Scaled job boards trade at 4–6× sales. The last Indian internship deal: about 2.3×.
| Company | Revenue (date) | Growth | Margin / profit | Multiple |
|---|---|---|---|---|
| Info Edge | TTM ₹3,473 Cr | +13.2% | Recruitment ≈59% | 22.2×* |
| Kanzhun / Boss Zhipin | TTM $1.28B | +14.1% | Adj. op. 43.8% | 4.0× EV/S |
| Recruit (group) | TTM ¥3.86T | +14.4% guide | EBITDA+S 26.1% | 6.1× |
| SEEK | FY26 ≈A$1,199M | +10% | EBITDA ≈44% | 4.3× |
| ZipRecruiter | TTM $452M ($905M FY22) | Q2 +5% | Adj. EBITDA 12% | 0.68× |
| Handshake | ≈$1.1B gross, ≈$450M net | +349% gross | Contractors 60–70% | 7.8× net, stale |
| Mercor | ≈$2B run rate, claim | Very high | n/a | 5× gross |
| WorkIndia | FY25 ₹78.7 Cr | +25% | Loss ₹23.06 Cr | ≈10.2× |
| Apna | FY25 ₹122 Cr | −4% | Loss ₹50.1 Cr | ≈81×, 2021 mark |
| Internshala | FY25 ₹44.1 Cr | +19.5% | Loss ₹11.0 Cr | ≈2.3×, unconfirmed |
Market caps 6 Nov 2026 (stockanalysis.com). Multiples DERIVED, mostly ignoring debt and cash. *Includes investment stakes, so overstates core. Refresh after ZipRecruiter (9 Nov), Recruit (11 Nov), Kanzhun Q3.
Revenue multiple, usable marks only
ZipRecruiter
0.68×
Internshala≈2.3×
Kanzhun4.0×
SEEK4.3×
Mercor5×
Recruit6.1×
Handshake7.8×
WorkIndia≈10.2×
Excluded: Info Edge 22× (stakes), Apna 81× (2021 boom mark).
Read-across
- Apna’s 2021 mark vs Internshala’s 2026 trade is the key underwriting fact: boom mark vs actual deal.
- Listed boards have de-rated despite strong margins: SEEK −46%, Kanzhun −25%, Info Edge −11%.
- Realistic exit lens for an Indian internship pure-play: 2–4× revenue DERIVED.
Competitor map · India
Nobody owns internships profitably in India. Naukri is the real copy risk.
| Player | Size (dated) | Who pays and price | Weak spot |
|---|---|---|---|
| Internshala upGrad 90% | FY25 ₹44.1 Cr, loss ₹11 Cr; 34M+ registered, ≈3M active applicants COMPANY; 6,171 live OBSERVED | Free: one post a month, first 20 applications visible. ₹2,999 per listing. Courses ₹499–₹2,999 on sale; Pro planned | Trust complaints, course conflict, thin outcome data |
| Unstop Coursera among investors | FY25 ₹33.1 Cr (+50.7%), loss ₹12.8 Cr; 32M+ users COMPANY | Employers quote-only. Pro price sources conflict (₹1,799–₹10,999) | Low-star reviews are mostly app bugs |
| Naukri Campus Info Edge | 4.5M+ student profiles COMPANY; recruitment ≈₹2,374 Cr FY26 | Hot Vacancy ₹1,650; Priority Applicant ₹432 a month (competitor blog) | Not intern-first. Highest copy risk: can bundle free |
| LinkedIn, Indeed | LinkedIn 167M India members PRESS; no India revenue found | Pay-per-click promoted jobs | Internships are not their focus |
| Apna | 5.9 Cr seekers, 7 lakh employers COMPANY; recruitment ₹57.1 Cr FY25 | Plans from ₹1,949 a month (competitor blog) | Blue-collar. Apna Safety is the trust benchmark |
| Instahyre, Cutshort | ₹18.9 Cr, profit ₹3.1 Cr; ₹2.7 Cr, profit ₹9.8 lakh | Cutshort: flat plans, no usage limits | Tech only. Proof small and flat can be profitable |
| Superset Great Learning | 600+ colleges; ₹10.8 Cr FY25 (single source) | Colleges appear to pay; no public price | Edtech-owned; parent in insolvency |
| HelloIntern | One category showed “1 open” OBSERVED | Free posting | Same “every employer reviewed” pitch, inventory appears thin. The claim is cheap; liquidity is hard |
| Let’sIntern, Twenty19 | letsintern.com serves crypto-casino content; twenty19.com didn’t respond OBSERVED | n/a | Hollowed out. Two cases, suggestive only |
Takeaway: Internshala and Unstop both lose money at ₹33–44 Cr revenue. The two profitable models are small, flat-priced and employer-paid. If upGrad also won Great Learning it would hold both Internshala and Superset; both upGrad and Manipal filed interest, and no 2026 outcome was found PRESS.
Prices via Internshala’s blog are competitor-published and need checking. Sources linked inline; see also YourStory and NewsBytes.
Internsive · Discussion memo
Competitor map · Government
The PM Internship Scheme is India’s best public funnel, and it’s now a live competitor.
PMIS funnel, candidates at each stage
Round 1Round 2
Offers
Accepted
Joined
Exited
60,866
71,195
28,144
24,638
8,760 (14.4%)
7,300 (10.3%)
4,702 (53.6% of joiners)
2,464 (33.7%)
- Accepting cost nothing, so it signalled little. About 70% of acceptors never joined DERIVED.
- Why they didn’t join (740 non-joiners surveyed, about 4% of them): 61% personal, including exams and other jobs; 16% location; 8% low pay; 8% poor communication. Most midway exits came within 15–30 days PRESS.
- Only 95 of 16,060 joiners had full-time offers by Dec 2025 PRESS. Don’t promise conversion.
- Round 3 looks better but is confounded: stipend +80%, shorter term, more companies and final-years all changed at once.
Other rails
| Scheme | Figures | Why it matters |
|---|---|---|
| PMIS Round 3 | ₹9,000 a month, 6–9 months; 30,000+ joined by 13 Oct 2026; next step undecided PRESS | Visible stipend anchor. Final-years eligible from Apr 2026 |
| AICTE portal | 4.7 Cr registrations, 79.6 lakh “opportunities”, 32.8 lakh “completed” SECONDARY | Not comparable; definitions look looser, unverified |
| NAPS | 1,95,939 establishments, about 26% active; 9.85 lakh engaged PRESS | Underused on the employer side |
| NATS | 5.23 lakh engaged; stipends reportedly ₹9,600–12,300, 50% reimbursed, capped ₹4,500 SECONDARY | Larger subsidy than NAPS’s ₹1,500 cap. Needs checking |
Budget signal
FY26 estimate ₹10,831 Cr; spend ₹64.91–73.72 Cr across two reports. FY27 estimate about ₹4,788 Cr could fund about 4.4 lakh intern-years, about 10× current joiners DERIVED. Largest single policy swing for us: offers close in December, joining in January, and the government “has not taken a final call”.
Sources: ThePrint Mar 2026 · The Wire · ThePrint Dec 2025 · ThePrint Oct 2026 · NITI via ThePrint · Tax Garden. Round 1 completed: 3,417. Source figures conflict; see appendix.
Internsive · Discussion memo
Competitor map · World
Abroad: the winners used campus software, public data or direct chat. Swipe is noise.
| Player | Figures (dated) | Lesson for Internsive |
|---|---|---|
| Boss Zhipin (China) | 70.2M MAU, 7.2M paying firms, Q2 2026 HARD; ≈₹17,500 per paying firm a year DERIVED | Possibly the closest template: low price, huge employer count, direct chat, resume on consent |
| Handshake (US) | Universities ≈$8,000 a year, employers from $15,000; core ≈$190M 2024, +6%; AI training ≈$1B gross PRESS | Cheap campus software forces supply on. University fees are acquisition, not revenue. The verified network was the option |
| Simplify (US) | 800K+ users COMPANY; 48,321 GitHub stars (7 Nov 2026) | A useful public list carries the product. Documented jumps came from LinkedIn and printed campus cards; mass email failed |
| Jobright (US) | 50,000 users with no marketing, June 2024 PRESS | Public government data turned into a filter for an underserved group |
| Parker Dewey (US) | $200–600 a project, 90% to student SNIPPET | About $30 per project to the platform. Needs institutional money |
| Forage (US) | 10M engagements by Aug 2025, after embedding in EAB software COMPANY | Campus software as channel. Outcome claims are self-selected |
| RateMyPlacement (UK) | 65,000+ student reviews SNIPPET | A niche review wedge survives beside big boards |
| JobTeaser (France) | ≈500 universities, 7 lakh students, Jan 2018 SECONDARY | Reached Handshake’s model separately: a sturdy mechanism |
| Welcome to the Jungle | $83M+ raised; flat revenue, not yet profitable, 2025 SNIPPET | Employer branding plus capital doesn’t guarantee profit |
| Shixiseng (China) | 1M MAU, 30%+ two-year retention, Dec 2017 PRESS | An internship vertical can reach coverage on little capital. No recent data |
| Prosple (Australia) | $6.3M revenue, Nov 2023 PRESS | Printed campus guides first, then digital. Already runs college portals in India |
| Sorce (US) | 850,000+ users (claim), Sept 2026; 3.01 stars on 250 reviews (competitor-written) | Student swipes, AI applies, no employer side. Noise, not a template |
Also checked: Bright Network (about £11.6M, aggregator), Wantedly (¥4,908M, +4%), CIEE Brazil (174,600 contracts). Internships.com is live under Chegg, Hired was absorbed by LHH, Forage went to EAB. Coverage is thin for Latin America, Africa and the Middle East.
Structural conflict
Incumbents earn from what our pledges attack. Copying would hurt them, but only some of them.
Why they may not fix it
- Internshala sells to the students it serves, and is now owned by a course seller. upGrad: FY25 revenue ₹1,569.3 Cr, net loss ₹273.3 Cr PRESS. The Ken frames the deal as an early-career funnel and traffic shortcut. The stated plan includes a paid candidate “pro” tier PRESS.
- Its own page shows the tension. A placement course says “GET HIRED” while stating it “does not guarantee a job or internship” OBSERVED.
- Its paywall feeds on volume. Only the first 20 applications show on a free listing OBSERVED. A cap on student applications would weaken that upgrade prompt [inference].
- Unstop sells courses and Pro beside hiring, and Coursera is an investor PRESS.
- Naukri, foundit and Freshersworld sell candidates a way up the queue COMPANY.
- Scoreboards embarrass paying customers. A listing seller has little reason to publish which employers pay late [inference].
Treat this as a hypothesis, not a measured fact. The most useful missing number is Internshala’s current training share of revenue: old reports conflict at about 65% and 25%.
What this does not protect against
- Naukri Campus could copy “no courses, no pay-to-rank” more easily, because its money comes from employers.
- Internshala’s installed base: 34M+ registrations, 13,000+ college partnerships per its FAQ, 40%+ tier-II/III users COMPANY.
- Internshala already has the regulator’s channel: a free course with AICTE, 600,000+ learners reported COMPANY.
- Features can be copied in a sprint. The argument is only that copying hurts current revenue, and current internal economics are unconfirmed.
Trust benchmark to beat
Apna Safety (Sept 2025): GST, PAN, CIN and Aadhaar checks plus a public recruiter lookup. Company-reported pilot: 146,000+ recruiters verified, 45% less scam exposure, not independently verified. Any trust claim we make will be compared with this.
One-snapshot warning
Internshala’s homepage on 7 Nov 2026 featured roles at upGrad Learnings and Vedantu inside sales OBSERVED. How slots are chosen is unknown, so don’t present this as a pattern.
Sources: MediaNama · The Ken (paywalled, opening only) · Inc42 · Internshala launchpad · Internshala FAQ · YourStory · Naukri FAQ · Internshala AICTE
Internsive · Discussion memo
User pain · Ranked
Trust and silence lead the complaints, but nobody has measured how common scams are.
| # | Pain | Evidence strength | Best evidence |
|---|---|---|---|
| 1 | Fake, fee-charging or misleading listings | Strong that it exists; prevalence unknown | AICTE terms reportedly bar charging students; IIT Roorkee warning; VTU students allege payment demands |
| 2 | Silence after applying, including free assignments | Many reviews on several platforms; no Indian survey | Internshala on Play: “very little transparency about what happens after applying” (26 votes) REVIEW |
| 3 | Paid courses, plans, refunds, sales calls | Many complaints; affects the paying minority | Keyword tally on next page; Naukri on Trustpilot REVIEW |
| 4 | Unpaid, low, delayed or changed stipend | One large vendor survey plus anecdotes | One in four undergraduates unpaid in 2024, up from one in eight COMPANY; denominator unclear |
| 5 | Distance, relocation, moving cost | Strong for PMIS only; generalising is inference | Ministry cites a 5–10 km limit PRESS |
| 6 | Spam calls after sign-up | Reviews on Naukri and Apna; the tally overstates it | Keyword tally on next page |
| 7 | Recruiter noise, AI-written applicants | One vendor study without sample size, plus viral stories | LinkedIn India, Feb 2026 COMPANY |
| 8 | Intern no-shows and early exits | Hard data only from PMIS | About 41% of joiners left early PRESS |
| 9 | Stale listings | Weak as a stated complaint; probably reported as “no response” | 1–2% of low-star reviews |
Colleges feel it too
VTU requires a six-month, nine-credit internship for at least 85,000 students a year, and its Vice-Chancellor says it is “impossible for everybody to get opportunity”. Its location app showed students apparently “interning” at bars and gyms PRESS.
Scams don’t only live on platforms
One Hyderabad student paid ₹1,300 for a “typing internship” advertised on Instagram, and offers spread through college WhatsApp groups PRESS. One case. It shows the mechanics, not the scale. A stricter marketplace alone won’t stop this. The “80–90% fake” claims are users’ impressions, not measurements.
The ranking is our synthesis: no survey ranks internship-platform pain in India. Reddit could not be read. Deccan Chronicle for the Hyderabad case.
Internsive · Discussion memo
User pain · Evidence
Each incumbent has a different weak spot: Internshala’s is trust and courses, Unstop’s is its app.
Share of written 1–2 star Google Play reviews mentioning each theme, Jan 2025 to early Nov 2026
| Theme | Internshala n=346 | Unstop n=230 | Naukri n=1,006 | Apna n=867 |
|---|---|---|---|---|
| Fake, scam, fraud | 30.3% | 2.2% | 14.6% | 26.9% |
| Pay, fees, course, refund | 28.6% | 2.2% | 23.4% | 15.8% |
| Spam, calls, ads | 10.1% | 2.6% | 29.4% | 16.8% |
| No response, ghosting | 11.0% | 1.7% | 11.3% | 4.7% |
| Assignment, free work | 4.9% | 0.4% | 0.0% | 0.1% |
| Bugs, login, OTP | 6.6% | 45.7% | 5.9% | 7.4% |
| Live Play rating, 7 Nov 2026 | 4.34 | 4.01 | 4.61 | 4.70 |
How not to misuse this table
- A rough regex count, not hand-coded DERIVED. It measures what unhappy reviewers write, not how many users are affected.
- “Spam” is inflated, because “calls” also matches “no calls”. “Assignment” also catches course assignments.
- Naukri and Apna hit a 750-per-star fetch cap, so their samples lean recent.
A warning from the best Indian RCT
Vocational graduates registered on a job portal were 9 percentage points less likely to be working 12 months later, because they raised their wage expectations. Those who got 17 more SMS alerts were only 4 points below control STUDY. A marketplace that over-promises can harm its own users.
Employer side
- 78% of organisations run internship programmes; only 16% convert more than 80% of interns (500+ HR leaders, self-selected) COMPANY.
- One founder reported 1,290 applicants, 83 at interview, none cleared the test. Viral and unaudited: an illustration, not a base rate.
- Employer reviews on Apna allege irrelevant profiles and unclear plan terms REVIEW.
Trustpilot is harsher but thin
Internshala 3.0 on 104 reviews, Naukri 1.6 on 199, Apna 1.3 on 45. Rannkly shows Internshala at 4.4 on 3,217. Neither is unbiased.
The selling point
Four pledges, each backed by a rule. A pledge with no rule is just another badge.
Student: “Checked company, paid stipend, always an answer, never a fee.”
Company: “Fewer applications, from students who chose you, at one flat published price.”
| Pledge | The rule that makes it true | Pain rank | Honest limit |
|---|---|---|---|
| Checked company | GST, CIN or PAN name match plus human review before publishing | 1 | “Checked” means the stated facts, not “safe”. Counsel to review the wording |
| Paid stipend | No unpaid listings; a floor set by policy (Decision 4) | 4 | No legal minimum for interns. Proof of actual payment arrives later, through the ledger |
| Always an answer | Employer must respond by a set day, else the application auto-closes, the student is told, and the employer’s public reply figure falls ASSUMPTION | 2 | Sometimes the answer will be “no reply from this employer” |
| Never a fee | No charges to students, no courses, no pay-to-rank; employers barred from charging | 1, 3 | Removes a revenue line that incumbents rely on (Decision 3) |
Why this line
- Maps to the top four complaints, which AI matching and swipe do not.
- Hard for a course-selling owner to say with a straight face, though Naukri Campus could say much of it.
- “Never a fee” is also a channel rule: Google for Jobs disallows postings that require payment from applicants HARD.
Why the company line
- Answers recruiter overload, and the flat price contrasts with Internshala’s 20-application cap.
- Cutshort shows flat, no-usage-limit pricing working in tech OBSERVED.
- “Students who chose you” is only true once signals exist.
Don’t say / what makes it empty
- Don’t pitch “they hide prices”. Internshala’s blog lists ₹2,999.
- No evidence yet shows students switch for trust, or that a badge changes behaviour.
- The pilot tests this before anything is built.
Production gap · Overview
Not close: 100–115 engineer-days, plus a company, a lawyer and a CA, and a person on legal clocks.
75 daysResearch floor: defensible launch, no chat, no money between parties, no resume builder. 8–10 weeks for two engineers.
+25–40 daysWiring demo screens to a database (15–25) and the resume builder (10–15) ASSUMPTION.
₹23,300/mo10,000 sign-ups, 500 companies: SMS ₹2,000, WhatsApp ₹4,071, company checks ₹4,185, Supabase and SES ₹13,050 DERIVED.
13–14 May 2027DPDP main duties start. MeitY floated cutting 18 months to 12; check the Gazette before launch.
The ten things that matter most
- Label the public demo as sample data, and remove false claims, now.
- Real accounts and a database in place of localStorage.
- Company verification before a listing goes live.
- 18+ gate; nothing stored for under-18s.
- Consent notice, ledger, withdrawal and export.
- Grievance officer, 7-day clock, 3-hour takedown inbox.
- Secure resume upload, with per-application release.
- Report button, takedown queue, admin panel.
- Reply rule with auto-close, which makes the pledge true.
- Server-rendered listing pages with JobPosting data.
Not in those numbers
- Lawyer, CA, translation and accessibility audit fees were not researched. Salaries are extra.
- Log storage in India, and whether Supabase Mumbai is available, are unconfirmed.
Three common errors to avoid
- “15-day grievance clock.” It has been 7 days since 20 Feb 2026, with 3 hours for government and court orders.
- “2 crore means Significant Data Fiduciary.” That is a retention threshold. No numeric SDF threshold exists.
- “Google can’t see our pages.” True today, because the localStorage demo is invisible to it.
Effort figures are research estimates, not sourced. Sources: DPDP Rules · Shardul Amarchand on IT Rules · MirvoLegal · Supabase
Internsive · Discussion memo
Fix and add · Foundations, demo, sign-up
First: a legal base, an honest demo and real accounts.
A. Before any code
- ADDBLOCKERCompany, PAN, GST, current account. 2–4 weeks.
- ADDLEGALTerms, privacy notice, employer agreement. 1–2 weeks.
- ADDLEGALNamed Grievance Officer: acknowledge in 24 hours, resolve in 7 days, monitored inbox for 3-hour orders. 1 day.
- ADDLEGALE-commerce grievance officer and nodal person, if anyone pays SNIPPET. One person can hold all these roles.
- ADDLEGALCERT-In contact, 6-hour incident runbook, clock sync. 1–2 days.
- ADDLEGALDLT registration before any SMS. Start in week 1.
B. The public demo
- FIXBLOCKER“Demo, sample data” banner on the live site now. Hours.
- FIXBLOCKERRemove all sample companies, internships, applicants and activity before real users arrive. Google lists fake jobs as violations.
B. On-screen claims that aren’t true
- FIXBLOCKER“Verified company” → a dated fact: “GSTIN active, name matched on 7 Nov 2026”.
- FIXBLOCKER“Free to apply”: only once enforced by attestation, keyword filters and takedown.
- FIXBLOCKER“Replies in about N days”: compute from real data, above a minimum sample of say 10 decided applications. Else “New employer, no reply data yet”.
- FIXBLOCKERMatch %: replace with plain reasons. A number with no outcome data is a claim you can’t defend.
- FIXBLOCKERActivity panel: real platform facts only.
- FIXNICEThe 73,051 list is an unfiltered AISHE directory export. The 2023-24 report counts 64,756 institutions. The 8,295 gap is unexplained, so say “from the AISHE directory”, not “every active college”.
C. First screen and student sign-up
- FIXBLOCKERlocalStorage → database, auth, sessions. Supabase Pro about ₹2,250 a month plus recovery add-on. 10–15 days.
- FIXBLOCKERMake last name optional. Many Indians use one name ASSUMPTION. Minutes. (This reverses your last request; your call.)
- ADDBLOCKERGoogle sign-in plus phone OTP, WhatsApp first (₹0.115 + GST), SMS fallback (≈₹0.15–0.25). Avoid Twilio at ≈₹7.50. Rate-limit OTP. 2–3 days.
- ADDLEGAL-202718+ gate. Penalty ceiling ₹200 Cr. Store nothing for under-18s, not even a waitlist email. 1 day.
- ADDLEGAL-2027Stand-alone consent, consent ledger, withdrawal, export, two-tier delete. 5–7 days.
- ADDLEGAL-2027Consent notice in English or any Eighth Schedule language. 3–5 days plus translator.
- ADDNICEVerification ladder: phone, email, “Verified at college” from a domain list, optional DigiLocker (≈₹2.49), ID review. Don’t mandate .ac.in. Don’t store Aadhaar.
- FIXNICEAsk less at sign-up, the rest at first application. Keep Guest.
Fix and add · Company, posting, listings, applying
Then: check every company before it goes live, and make each listing truthful and findable.
D. Company sign-up and verification
- FIXBLOCKER“Internship lister” has no checks today. Add a ladder and publish only after verification. 4–6 days plus a review queue.
Tier 1: work email matched to website, plus GSTIN or CIN name match. Tier 2: PAN plus bank penny drop. Tier 3: human review for sole proprietors, startups without GST, on-site roles, high stipends and scam keywords. - Cost: GSTIN ₹0.90 + CIN ₹2.99 + PAN ₹1.49 + bank ₹2.99 = ₹8.37 per company COMPANY. Budget ₹10–25.
- ADDBLOCKERNew employers get no bulk candidate data. Resumes released per application only.
- Why legally: in interim orders in Reliance v. Quikr and OLX (Delhi HC, 2020), the court rejected “cannot verify” on fake-job ads. Not a final liability ruling.
- Why not enough: a student built a fake company on Handshake in under two hours in 2018 PRESS.
E. Posting an internship
- FIXBLOCKERRequire a class: curriculum-linked, standalone paid, or registered apprenticeship. A paid intern doing productive work can be treated as an employee.
- FIXBLOCKERRequire stipend, work address, hours and expiry date.
- ADDBLOCKERAttestations: no fees, real role, stipend, POSH committee contact, night hours. POSH covers unpaid interns; the duty stays with the host. 1–2 days.
- ADDBLOCKERPre-publish flags: fee requests, deposits, Telegram- or WhatsApp-only contact, unrealistic pay. Don’t build on Perspective API, which closes 31 Dec 2026.
- ADDNICEAssignment rule: paid, time-capped or answered in a set time.
- ADDNICEFlat published employer price, no cap on visible applications.
F. Browsing and the listing page
- FIXBLOCKERServer-rendered public pages: one JSON-LD block per listing with
INTERN, real stipend, expiry, removal when filled, Indexing API and sitemaps. 5–7 days, free. - FIXBLOCKERNever put the description behind login. Empty city or college pages stay
noindex. - FIXBLOCKERPerformance budget: about 0.3 MiB of JS on the critical path, tested on a Galaxy A-class phone at India’s P75 of 6.2 Mbps SECONDARY. 3–5 days, with the PWA.
- ADDNICECommute-radius filter. Internshala shows city and WFH filters but no radius, logged out OBSERVED.
- ADDNICEPWA first. A Play wrapper needs report and block, and personal developer accounts need 12 testers for 14 days, so register as an organisation. 2–3 days.
G. Applying
- FIXBLOCKERSlide-to-apply needs a keyboard and screen-reader path. The disability commissioner has directed private apps such as Practo to become accessible and fined establishments in 2025 SECONDARY. 3–5 days plus audit.
- ADDBLOCKERResume upload: private bucket, signed URLs, 2–5 MB, PDF and DOCX only, scanned and quarantined until clean, every recruiter download logged. 3–4 days.
- ADDNICEOne or two screening questions per listing. Frictionless swiping reportedly made poor-fit applications too easy SECONDARY.
- ADDNICEResume builder: see its own page.
Sources: Message Central rates · S.S. Rana · Inside Higher Ed · CalcGuru · LawBhoomi · Google JobPosting · Infrequently Noted · Play UGC · AZB
Internsive · Discussion memo
Fix and add · Replies, pipeline, safety, data, money
Finally: make “always an answer” real, staff the safety clocks, and keep stipend money out.
H. Tracker, replies, notifications
- FIXBLOCKERTracker states from real employer actions.
- FIXBLOCKER“Passing tells the student straight away” plus Undo can’t both be true. Hold each pass for a short window, then send; never un-send.
- ADDBLOCKERReply rule with auto-close.
- ADDBLOCKERNotifications: WhatsApp utility ₹0.115 + GST, email $0.10 per 1,000. Marketing templates cost 7.5× more, keep opt-in.
- ADDBLOCKERINR WhatsApp billing: delivery stops for eligible non-INR accounts from 1 Jan 2027.
- ADDNICEConfirmation before start, and a check-in within 15–30 days, where PMIS exits clustered.
I. Pipeline and swipe
- FIXNICEPipeline as default, swipe as optional triage. Drop “It’s a match”. Keep no photos.
- FIXBLOCKERNon-gesture alternatives for all three swipes.
- FIXBLOCKERScores may sort, never auto-reject.
J. Messaging
- ADDNICESkip at launch. 6–8 days later; Play then requires block and report. A known parity gap: Internshala includes it free, and Boss Zhipin’s direct chat is the best commercial evidence here.
K. Trust, safety, admin
- ADDLEGALReport button everywhere, takedown queue with 3-hour, 36-hour and 7-day timers. 8–12 days with admin.
- ADDBLOCKERAdmin panel: roles, review, suspend, grievance inbox, breach log. Low-code is fine at first.
- ADDBLOCKERBot check and per-account, per-IP limits on sign-up, OTP, apply, posting. 3 days.
- ADDBLOCKERHuman review of every listing during soft launch.
- ADDNICECross-campus flagging: one college’s report protects all.
L. Data and security
- ADDLEGALAudit logs: 180 days in India (CERT-In), one year from May 2027. Supabase Pro keeps 7 days, so export them. 3–4 days.
- ADDLEGAL-2027Encryption, access control, monitoring, processor clauses. Penalty ceiling ₹250 Cr.
- ADDLEGAL-2027Breach runbook: tell users and the Board without delay, full report in 72 hours, no materiality threshold.
- ADDLEGAL-2027Tested restores, published privacy contact, grievance time of 90 days or less.
- ADDBLOCKERError tracker, uptime check, on-call alias.
- ADDBLOCKERNotice says resumes go to the named company on apply. LLM vendors need a processor contract.
M. Money and tax
- FIXBLOCKERKeep the rule: no stipend money through Internsive at launch. Holding third-party funds looks like payment aggregation: RBI authorisation and ₹15 Cr net worth. Razorpay Route needs ₹40 lakh+ turnover. Refundable sums may raise deposit-scheme questions.
- ADDNICEMerchant-model gateway for employer fees: Razorpay 2% + GST on the fee. 2–3 days.
- ADDLEGALGST registration once applicable, likely 18%. CA to decide, along with TDS.
- ADDBLOCKERWhen charging: plain refund terms, no auto-renewal traps. A consumer commission ordered Shine to refund ₹79,751 SNIPPET.
N. Measurement
- ADDBLOCKERTrack joined, completed, reply time and stipend-paid confirmation, not listings or applications.
Targets (assumptions; no public benchmark exists)
- 5+ screened relevant applicants in 7 days on over 70% of listings
- First relevant application in under 48 hours
- Offer-to-join above 50% (PMIS: 10–14%)
- Employer repeat posting above 40% at 6 months
Your idea · Resume builder
Right instinct, wrong billing: worth having, cheap to build, impossible to charge for, not a moat.
Why it’s table stakes
- Internshala’s is already free: PDF download, upload parsing, AI help, scoring, feedback OBSERVED.
- Naukri: 3 AI attempts and 3 templates free, more with Pro SNIPPET.
- Abroad: Rezi $29 a month; Resume.io €24.95 per 4 weeks after a €2.95 trial, flagged for auto-renewal SNIPPET.
Why still build it
- Onboarding already holds most of the content.
- About 10–15 engineer-days ASSUMPTION.
- First-job students have thin records and need a start.
Upload is a launch blocker. The builder is not.
What the evidence says (US, mostly vendor-run; no Indian recruiter study)
- Editing help works. Algorithmic writing help led to 8% more hires at 10% higher wages, across nearly 500,000 jobseekers. It fixed errors and readability; it didn’t ghostwrite STUDY, SNIPPET.
- “ATS auto-rejects you” is largely a myth, on limited evidence. 92% of 25 US recruiters said it doesn’t reject for formatting. Knockouts do reject. Counterpoint: 21% of 948 US leaders let AI auto-reject at some stage SNIPPET.
- Managers say they dislike ghostwriting but detect it poorly. 80% say they can often tell; only 33.5% picked the AI resume against a 25% chance baseline SNIPPET.
- AI can invent facts. Users report Jobright’s tool adding skills and credentials SNIPPET.
Build
- 1–2 single-column templates, selectable-text PDFs, from the profile.
- Import by parsing; student reviews every field.
- A checker showing what a recruiter’s system would extract, plus missing skills for a chosen listing.
- AI that edits the student’s own facts and flags invented claims.
Don’t build
- “ATS score”: mostly marketing, possibly a misleading claim.
- A paid tier, trial auto-renewal, or full ghostwriting.
The khurafati version
Every verified internship prints as a resume line with a public check link. Every PDF sent to any recruiter then advertises the record. Jobseekers with credibly shareable results gained employment and earnings STUDY. It needs the ledger first.
Sources: Wiles, Munyikwa, Horton · Enhancv · Lenz · Resume Genius · HRO Today · Careery · Carranza et al., AER 2022 · ToolRadar · Dupple
Internsive · Discussion memo
Khurafati moat ideas · The book
Bet on three: free capped signals, a stipend ledger, and a Trojan logbook for colleges.
| # | Idea | Evidence | Small-team fit | Why an incumbent resists | Option it opens |
|---|---|---|---|---|---|
| 1 | Free capped signals | Strong elsewhere | Very high | Cuts the volume they sell | Match data, ranking |
| 2 | Verified stipend ledger and employer scoreboard | Mixed | High | Embarrasses paying customers | Stipend index, credentials, AI labour |
| 3 | Free college logbook | Strong elsewhere, mixed in India | High in one state | Discourages paid certificates | Faculty distribution, DigiLocker links |
| 4 | Scam Check for any offer | Need strong, tool untested | High | Little gain off their own shelf | Free acquisition, scam data |
| 5 | Public India internship list | Mixed | High | Won’t list rivals or direct links | Habit, backlinks, employer proof |
| 6 | Commute-first supply and honest odds | Mixed | High, slow | Low odds deter course sales | Dense local network |
| 7 | Stipend-backed badge | Weak for hiring | Medium | Pre-funding cuts listings | Paid-stipend data, payments |
| 8 | Paid micro-trials | Strong academically | Medium | Conflicts with course funnels | Work record, conversion fee |
| 9 | Voice answers | Blue-collar only | High | They don’t: it’s a feature | Screening dataset |
| 10 | Family share card | Mixed | High for the card | Safety data hurts payers | Women’s trust |
Later, exercised only on evidence
- NATS paperwork layer: handle apprentice onboarding, pitching a reported 50% reimbursement capped at ₹4,500 SECONDARY. About 26% of NAPS establishments are active. Check the rule first.
- AI training work: Indian-language tasks from verified students. Handshake proved it for experts; Indian rates are far lower, and contractors have reported pay disputes. Only the verification link lasts.
- Curated foreign-startup batch: 10–30 hand-picked startups, prepaid stipend, paid directly to the student. No purpose code names a stipend, so get a CA opinion first.
- Campus collective unlock: a whole college unlocks something together. No outcome data. Small test only.
How to read the book
- The ranking is judgement, using your lens: moat as optionality, things that compound and that incumbents can’t copy without hurting how they earn.
- Nothing here has been tested in an Indian internship market.
- Most protection is counter-positioning, not a technical barrier. Features copy in a sprint; a two-year record doesn’t.
- Naukri is the exception. It earns from employers, so most of this doesn’t bite it.
- Spreading a tiny team across all of these is the main stated risk.
Our additions to the research ranking: Scam Check and the open list were raised but unranked; the stipend-backed badge moved down because of its legal gate.
Moat ideas · Top two
Signals are the cheapest strong bet. The stipend ledger is what compounds.
1
Free capped signals: “I actually want this one”
- Mechanism: 3–5 free priority signals a month that can’t be bought. Employers see signalled applicants first and see how many they got. Coach students to signal realistic employers.
- Precedent: economists’ job market, two signals since 2006–07 (signals to much higher-ranked departments were mostly wasted). US residency 2023–24: 4.7–7.4× interview odds with few signals, observational. Roses in a 613-person dating experiment raised acceptance.
- Evidence: strong elsewhere; no Indian internship test.
- Main risk: employers ignore signals, an outcome theory allows. Or universal use makes it a tax: in orthopaedics nearly all applicants signalled, and signalled programmes took about 79% of interview offers.
- Why incumbents resist: they sell reach and volume, and a cap cuts the number they show. They could still copy it in a sprint.
- Rule: free, capped and never for sale. Paid signals turn an honest message into a wealth message.
Sources: AEA · Academic Medicine · Lee and Niederle · Coles et al. · AAOS 2026
2
Stipend Truth: a paid-stipend ledger and public scoreboard
- Mechanism: after joining, both sides confirm the stipend was paid, with optional proof. No money passes through us. Students see city and field stipend tables only after contributing their own. Employers get numeric public facts: reply time, paid on time, hires.
- Precedent: Glassdoor’s give-to-get. Levels.fyi, begun in 2017 on Google Sheets, verifying with offer letters and payslips SNIPPET.
- Evidence: mixed. Good pattern, mostly company-sourced numbers. Data doesn’t guarantee independence: Glassdoor was folded into Indeed in 2025.
- Main risk: empty until dense. Takedowns: Delhi HC ordered posts against PhysicsWallah removed in Jan 2026, while another bench dismissed LawSikho’s suit over criticism with ₹1 lakh costs. Numbers look safer than opinion. Lawyer to confirm.
- Why incumbents resist: they can publish listed stipends, not paid ones, without a proof trail, and the data embarrasses paying customers.
- Rule: numeric, one city first, no cell published below a minimum sample (research example: 40).
Sources: Glassdoor · Levels.fyi · HR Dive · Bar and Bench · Verdictum
Moat ideas · Distribution
Three crafty ways to get students and colleges without paying for them.
3
The Trojan logbook
- Mechanism: a free tool where a mentor signs off weekly, the employer is checked against GST or MCA, and it exports the NAAC 1.3.3 file and registrar’s credit list. Faculty become free distribution.
- Precedent: Handshake’s cheap campus system at about $8,000 a year secured supply first. VTU: 85,000 students a year need a nine-credit internship; about 34 students per registered company DERIVED, rough.
- Evidence: strong elsewhere, mixed in India. No source shows colleges paying. Superset’s ₹10.8 Cr suggests the money isn’t here.
- Risk: slow; big universities build their own. If it becomes a certificate mill, it destroys the brand.
- Tension: UGC’s 60–120 hour credit internship has no stipend rule, while we pledge paid listings. The logbook can record unpaid credit work without the marketplace listing it.
- Fit: 5–10 autonomous colleges in one state. Free forever.
4
Scam Check
- Mechanism: forward any offer from WhatsApp, Instagram or email. It checks the company against GST and CIN, looks for fee demands, and answers in plain language. Free, no sign-up, reaches college groups at the moment of fear.
- Precedent: Apna Safety’s public recruiter lookup COMPANY.
- Evidence: need is strong; the tool is untested and prevalence unmeasured.
- Risk: a false “looks fine” harms a student. Naming a real company invites defamation, so report facts (“GSTIN not found”), not verdicts. WhatsApp bars general-purpose AI assistants since 15 Jan 2026, so keep it narrow.
- Why incumbents resist: little gain from checking offers off their own shelf. Copyable, though, and Apna is close.
- Fit: high. Reuses company checks at about ₹8.37 a lookup; needs abuse limits.
5
The open list
- Mechanism: a free community list of good roles with direct company links, on GitHub and a WhatsApp channel. Each row has “apply direct” and “on Internsive”, with tracking tags that show up in the employer’s own data.
- Precedent: SimplifyJobs list: 48,321 stars, 3,269 forks, 7 Nov 2026 OBSERVED.
- Evidence: mixed. No source measures how many users the list brought.
- Risk: upkeep and stale rows. Unverified roles dilute “checked company”, so keep it separate. Never scrape logged-in competitors: hiQ ended in a $500,000 consent judgment.
- Why incumbents resist: a listing seller won’t publish rivals and direct pages.
- Fit: high. Start by hand. Phone-first students probably favour WhatsApp, though that’s unsourced.
Sources: Sacra · Deccan Herald · NAAC file example · Careers360 on UGC · Business Today · TechCrunch · Privacy World
Internsive · Discussion memo
Moat ideas · Rounding out the book
Five more worth keeping on the watchlist. Useful features, weaker moats.
| # | Idea | Mechanism | Closest precedent | Main risk and verdict |
|---|---|---|---|---|
| 6 | Commute-first supply and honest odds | Radius is the first filter. Recruit employers within 5–10 km of partner colleges. Show real response rates for this college, field and city. | PMIS’s 5–10 km limit. The Kelley RCT on realistic information. 59% of Internshala’s internships were in three metros in 2024 COMPANY; 70% of colleges sit outside the top 50 districts HARD. | Slow, unglamorous sales; tier-2 employers may be scarce; low odds deter sign-ups. A local employer base is what can’t be copied quickly. |
| 7 | Stipend-backed badge | Optional ranking benefit. Start with non-custodial proof, which needs no licence. Only later pre-load a first-month stipend through a licensed partner. | Upwork’s paid Connects HARD. No hiring marketplace with a refundable employer bond against ghosting was found. | RBI and deposit-scheme rules. ₹10,000 via Route costs about ₹248, and Route is closed below ₹40 lakh turnover. Only after counsel. Never from students. |
| 8 | Paid micro-trials | A one-week paid trial for a few thousand rupees, a shareable evaluation, and a short live call to defend the work. | Pallais randomly hired 952 inexperienced workers for 10-hour jobs; later earnings roughly tripled STUDY. Parker Dewey is the commercial version. | Free-labour optics and AI-assisted work. Skills-based hiring changed outcomes in fewer than 1 in 700 hires across 11,300 roles. Sell a lower-risk trial, not a philosophy. |
| 9 | Voice answers | Replace the cover letter with a two-minute voice answer in the student’s language, plus a short transcript for the employer. | Apna’s AI calling agent; “50% faster” is unaudited. Sarvam ₹3.50 a minute COMPANY, about ₹17.50 for a five-minute screen DERIVED. | Thin evidence for students, and no language-barrier data. Consent duties on voice data. A feature, not a moat. |
| 10 | Family share card | One tap sends a card in the family’s language with the checked employer, address, commute and mentor. Student’s choice only. Women’s reports shown only in aggregate. | Delhi women chose lower-quality colleges for safer travel STUDY. A family video raised take-up among 1,022 UP women, but employment after 13 months didn’t change STUDY. | Stalking and paternalism: never automatic, never “safe” as a guarantee. The card is cheap; the safety data comes much later. |
Sources: ThePrint · Careers360 · Upwork 10-K · Pallais · Harvard · Silicon India · Sarvam · Borker · McKelway
Internsive · Discussion memo
Ideas to kill
They sound clever. The evidence, the law or the selling point says no.
- 73,051 indexed college pages. Thin template pages match Google’s doorway and scaled-abuse examples HARD. The list is a public export, so it isn’t proprietary. Index a college only with unique verified data, starting with 50–200.
- Swipe as the product’s identity. Job seekers swipe (Jobr claimed 100M+ by 2016). No evidence of recruiters adopting and paying at scale. Switch and Shapr are gone or reportedly closed. Absence of evidence isn’t proof, so keep it as an optional view at most.
- Any charge to students, including a “refundable” deposit. “Pay to get an internship” is the scam, it disqualifies listings from Google for Jobs, and promising a refund carries deposit-law risk. Capped signals are the non-monetary cost.
- Paid signals, featured applicants, pay-to-rank. They break the selling point.
- Holding stipends in our own account. See the money section.
- Refundable employer bond at launch. No precedent, untested legality. Hold under idea 7.
- Course sales or paid certificates. Reintroduces the conflict. Credly’s roughly 15× price came with brand-name issuers; an unknown startup’s certificate has no such pull.
- “ATS score” and AI ghostwriting. See the resume page.
- AI auto-apply. Adds to the noise we sell against. A competitor-written sample of 250 Sorce reviews averaged 3.01 stars.
- Video resumes. TikTok Resumes was a short 2021 US pilot; critics warned of appearance bias. No adoption at scale found.
- User equity now. Securities limits weren’t researched. Urban Company’s ₹150 Cr partner plan didn’t prevent a worker campaign at its IPO. Equity doesn’t make up for poor terms.
- Large referral bounties and ambassador pyramids. In a 238-shop randomised rollout, larger bonuses raised referral quantity and lowered quality STUDY. Pay small amounts, only after a paid four-week outcome.
- Sign-up or data built on a rival’s API. One vendor blog says LinkedIn’s lockdown hurt Jobr. Scraping also carries contract risk.
- A Chrome extension as the main channel. Students are probably phone-first (unsourced), and platforms can act against extensions. Consider it after the list.
- A national launch or a large raise. A national cost base needs more revenue than Internshala has after 15 years.
Sources: Google spam policies · GSQi · TechCrunch · Jobtrain · Inside Higher Ed · WorkLife · MediaNama · Friebel et al. · Chrome policy
Internsive · Discussion memo
Catalysts
PMIS’s next step in December–January is the biggest swing. Summer hiring is the natural pilot window.
| Date | Event | Why it matters |
|---|---|---|
| 9 and 11 Nov 2026 | ZipRecruiter and Recruit results; Kanzhun Q3 pending | Refresh the comparables |
| 13–14 Nov 2026 | DPDP consent-manager provision starts; any 12-month acceleration would land around here | Check the Gazette this week |
| 3 Dec 2026 | Bikhchandani becomes Info Edge Group CEO | Possible change of stance on Campus and freshers |
| Dec 2026 – Jan 2027 | PMIS offers close in December, joining in January, next step undecided | Largest single policy swing |
| 31 Dec 2026 | WhatsApp INR migration deadline; Perspective API closes | Don’t build on stale plumbing |
| Jan – Apr 2027 | Likely non-elite summer internship hiring run-up ASSUMPTION | Natural pilot window. Check with placement officers |
| About 1 Feb 2027 | Union Budget, by convention | PMIS allocation signal; FY27 estimate was ₹4,788 Cr |
| 13–14 May 2027 | DPDP main duties start; s.43A of the IT Act repealed | Consent, security, breach, children |
| Jul – Aug 2027 | Elite campus internship recruiting opens, on older calendars | Only relevant if targeting IITs |
| Aug 2027 – Feb 2028 | upGrad’s 18–24 month revenue plan for Internshala matures | Shows whether the owner pushes Pro and courses harder |
| Undated | Think & Learn insolvency outcome | Could put Superset with upGrad |
Sources: DPDP Rules · S.S. Rana · Moneylife · ThePrint · Meta · Perspective · SAM · IIT BHU · Entrackr · YourStory
Internsive · Discussion memo
Risks
What would prove this thesis wrong. Thresholds are discussion anchors, not findings.
| Risk | Evidence | What would prove us wrong |
|---|---|---|
| No liquidity | No public benchmark. Beachhead needs about 800 employers and 2,250 active seekers, about 19% of one 150-college district’s seekers DERIVED | Under half the liquidity targets after two pilot cycles |
| Employers won’t pay enough | Internshala earns about ₹980 per employer ever acquired; free posts exist on several boards | Under 1 in 5 pilot employers pay anything for a second listing |
| Trust rules shrink supply | Most employers are small and listings are cheap | Under half of genuine employers complete checks, or a paid floor removes most listings |
| Employers ignore signals | Theory allows this outcome | No material lift in replies. The logbook becomes the main bet |
| Students don’t care | High store ratings; low-star reviewers are a minority | Students given both options still spray, and don’t cite trust when asked |
| Incumbent copies or bundles | Naukri has about 59% recruitment margins; upGrad can subsidise | Internshala launches capped signals or reply metrics and they work. The moat then rests on the ledger alone |
| PMIS full rollout | FY27 estimate could fund about 4.4 lakh intern-years DERIVED | Rollout covers all students at ₹9,000+ and employers migrate |
| AI cuts seats, enrolment falls | Stanford: ages 22–25 in exposed jobs about 19% below counterfactual, causation unproven. Yale: no broad disruption. TCS: about 25,000 fresher offers vs 44,000. No causal India study | Non-IT intern demand falls too. The non-IT focus exists to hedge this |
| A real harm on the platform | Boss Zhipin 2017: reports say a graduate was lured through a fake listing and later died, with a lawsuit and regulators summoning the company. Police reportedly later treated it as accidental drowning; court outcome not found SNIPPET | Any serious incident under a “checked” badge. Plan the response now |
| Legal exposure | Interim fake-job orders and defamation takedowns. No final ruling against a job portal found | Counsel says scoreboards aren’t defensible. Idea 2 shrinks to private data |
| Base rate | Internship-only boards tend to sell cheaply or hollow out | Plan either to widen beyond internships or to sell. Say which up front |
| Founder capacity | 100–115 engineer-days, plus human review, employer sales and a legal role on 24-hour clocks | One person can’t keep the 3-hour and 24-hour clocks and also sell |
Staged plan
Six bets, each sized to lose little, each with a condition that stops it.
| Bet | What we do | Stop or rethink if |
|---|---|---|
| A. Is the pain worth switching for? 2–3 weeks, founder time, no code | Hand-code 100 low-star reviews per app. Read Reddit in a browser. Walk through rivals logged in. Audit 100–200 live listings with public checks only, no fake applications. Interview 20 small employers and 5–10 placement officers. Take the unsafe claims off the demo. | Under half of employers accept checks before publishing. Under 1 in 5 would pay ₹1,000–3,000 a listing. Placement officers see no need. |
| B. Do signals and the answer rule change behaviour? 6–10 weeks, no-code | One city, one field, 2–5 colleges, 20–40 hand-checked employers. Forms and WhatsApp, run by hand, 18+, no money moving, a lawyer’s privacy notice. Half the listings get signals. Publish the open list. | Under half the liquidity targets after two cycles. No lift from signals: drop idea 1. Employers refuse to confirm stipends: idea 2 weakens. |
| C. Legal-minimum product in one state 100–115 engineer-days, ≈₹25,000 a month plus lawyer and CA | Only if B passes. Soft launch, human review of every listing, pipeline as default, no chat or money movement. | Review time per listing exceeds what it earns. Genuine employers fail checks at high rates. Reply rates fall once you stop chasing by hand. |
| D. Will employers pay? First paid season | One flat published price, tested inside ₹1,000–3,000, no visibility cap. Try a flat season plan with repeat employers. | Repeat paying under 20% at 6 months (target 40%+), or 390–1,300 payers looks out of reach. Then colleges as payers, or sell. |
| E. Does the ledger compound? Alongside C and D | Two-sided stipend confirmation, give-to-get tables, public scoreboard after lawyer review, verified resume lines. | Under 20% of joined interns have two-sided confirmation (target 50%+), or counsel won’t support a scoreboard. |
| F. The logbook One semester, 5–10 colleges, free | Only once real supply exists: a logbook without placements is paperwork. | Fewer than 2 colleges active after a semester. Requests for certificates without real work: stop at once. |
Options, exercised only on a trigger
| Stipend-backed badge | Turnover clears ₹40 lakh, counsel signs off, students clearly prefer backed listings |
| Direct chat | Employers ask for it, and block and report are built |
| AI work in Indian languages | A real buyer appears, with written appeal rights and fixed pay dates |
| NATS paperwork · DigiLocker issuer | Larger employers ask and rules are confirmed · Colleges are on the logbook |
All thresholds are assumptions. A small pilot gives direction, not proof. Nothing past Bet A should start without your decision.
Internsive · Discussion memo
Decisions needed · 1 of 2
Seven calls that shape everything else, most urgent first.
| # | Decision | Recommendation | Cost of choosing otherwise |
|---|---|---|---|
| 1 | Evidence or build first? | Bet A. 2–3 weeks | Risk spending about 5 engineer-months on an untested thesis |
| 2 | Pull false claims from the live demo? | Now, plus a sample-data banner | Fake “Verified” badges stay on a public URL |
| 3 | Ever take money from students, or sell courses? | Never, and say so publicly | Loses the selling point and the conflict argument. In exchange, employers must carry the whole business, which hasn’t yet worked at scale in India |
| 4 | Paid-only? What stipend floor? | Paid-only with a low floor at first, set after employer interviews. The logbook can still record unpaid credit work | A high floor kills small-employer supply. No floor empties “paid stipend”. ₹9,000 is PMIS, not a legal minimum; ₹12,300 is a graduate-apprentice reference, not an intern rule |
| 5 | Which city and field? | Where you can physically reach 40 employers, in a non-IT field. Karnataka has the best-documented shortage and Bengaluru’s 89% fresher intent, but VTU has its own portal | A distant city rules out the in-person sales the pilot depends on. A tech-first pilot meets the weakest demand and strongest rivals. No tier-2 calendar data exists either way |
| 6 | Launch age | 18+ only | Parent flow is 8–10 days plus legal; no ad-style personalisation for that cohort; ₹200 Cr penalty ceiling |
| 7 | Handle stipend money at launch? | None. Non-custodial proof first | RBI, deposit-scheme and employer-status risk |
The two that need an answer first
Decision 1 and Decision 2. Everything else can wait for the discussion, but the public demo currently shows “Verified company” and “Free to apply” on made-up listings.
Sources: People Matters (fresher intent, not intern intent) · Concall summary (IT billings)
Internsive · Discussion memo
Decisions needed · 2 of 2
Seven more, on what we publish, what we bet on, and what the end game is.
| # | Decision | Recommendation | Cost of choosing otherwise |
|---|---|---|---|
| 8 | Publish listings before or after checks? | After | Before gives faster supply, but a weaker diligence defence and the Boss Zhipin failure mode. After means a slower start |
| 9 | Keep the recruiter swipe? | Optional view, pipeline as default | The evidence doesn’t support recruiters adopting it. It may still help demos and small-firm owners, which is untested |
| 10 | Resume builder timing | After core, free, no ATS score; upload at launch | Spending 10–15 days early delays what’s legally required, for no differentiation |
| 11 | Which moats to bet on? | Signals plus stipend ledger, logbook third | Spreading a tiny team across mechanisms is the main stated risk |
| 12 | Publish facts about employers by name? | Numeric only, after lawyer review | Defamation exposure if loose; a much weaker moat if never public |
| 13 | What is the end game? | Decide openly. Small and profitable like Instahyre, sell to a larger owner, or widen beyond internships. The base rate favours widening or selling; 2–4× revenue is the realistic price lens | Leaving it unsaid risks building for a venture outcome the numbers don’t support |
| 14 | Who holds the legal roles, and what budget? | Get quotes during Bet A. Name one person for the grievance and CERT-In clocks | The 3-hour and 24-hour clocks need a real person watching |
Note on recent builds
This memo argues against three things we shipped this week: “It’s a match” framing, swipe as a headline feature, and required last name. Each is reversible and was built on your direction. They’re flagged here as open calls, not undone.
Appendix
What we don’t know, and where the sources disagree.
Known evidence gaps
- Reddit is unreadable. The largest hole in the student evidence; thread counts don’t exist.
- Logged-in behaviour is unverified. Gated features, application limits, chat and what Pro unlocks are unknown.
- No measured scam prevalence, and no Indian figure for fake certificates.
- No Indian test of any moat mechanism.
- No private-sector data on Indian intern offer-to-join, no-shows, exits or conversion.
- Internshala’s revenue split and paying-employer count are the two most useful missing numbers.
- FY26 filings weren’t found; no MCA filing was opened.
- No India revenue for LinkedIn, Indeed or foundit, so the market total is soft.
- Tier-2/3: no city-tier data in AISHE, no hiring calendar for non-elite colleges, no TPO survey.
- Language: no data on whether English-only platforms exclude applicants.
- Competitor prices mostly came through Internshala’s blog.
- Legal primary texts not read: RBI Directions PDF, UGC and MCA notifications, Lok Sabha replies. Lawyer, CA and audit fees weren’t researched.
- The Play tally isn’t hand-validated.
Main source conflicts (20 logged in the full memo)
| Topic | Versions |
|---|---|
| Internshala and unpaid roles | Says it bans them except NGOs and niche profiles, vs only recommends paid. Unresolved |
| Internshala growth | 30–40% (Screwvala) vs 19.5% FY25. We use 19.5% |
| upGrad deal | “90% stake” vs “90% of consideration by swap”; “for” vs “valued at” ₹100 Cr. Unconfirmed |
| PMIS offers, Rounds 1–2 | About 1.65 lakh vs 1.32 lakh. Joined rate 9.7–12.2% |
| PMIS exits | 6,618 (41%), Dec 2025 vs 7,094 (44%), Jan 2026 |
| Average stipend | ₹8,000 (Internshala), ₹25,432 (Indeed), ₹21,500 (job board). All means, each on its own mix. No medians |
| Handshake recruiting | About $190M (2024), $200M (Aug 2025), $150M (Apr 2026). Different methods |
| Handshake suit | Boggs v. Stryder Corp. per a docket summary, unverified. Not relied on |
| VTU companies | 2,500+ vs 14,000. We use 2,500+, so “34 per company” is rough |
| Simplify scale | 30M, 70M+, 100M+ or 200M applications. Unaudited, not used |
Full memo: every source, all 20 conflicts, complete arithmetic.