Problem
Over 13 lakh candidates appear for UPSC Civil Services each year, with an additional 25+ lakh writing state PSC exams. Most aspirants are in Tier-2/3 cities where access to quality mains answer-writing feedback is either unavailable or costs ₹1.5–2L/year at coaching institutes. Written answer quality is the single biggest differentiator between GS prelims passers and interview candidates, yet most aspirants receive zero structured feedback on their daily practice.
Solution
A web and mobile app where aspirants photograph or type their UPSC GS mains answers, and an AI evaluator — fine-tuned on UPSC model answers, DOPH guidelines, and past toppers' annotated scripts — returns structured feedback within 60 seconds: content coverage score, keyword density, structure rating, and a rewritten model paragraph. The v1 covers all four GS papers and Essay with a daily answer challenge feature. Payment via UPI; subscription unlocks unlimited evaluations and a 90-day progress tracker.
Why Now
India's AI-native startup wave (Inc42, 2026) and the demonstrated appetite of investors like Accel and Prosus hunting "off-the-map" India niches (TechCrunch, March 2026) signals that AI applied to non-glamorous but massive Indian verticals is an investable thesis. The LLM cost curve has dropped ~80% since 2024, making per-evaluation costs under ₹2 commercially viable at ₹999/month price points. UPSC coaching remained stubbornly offline and expensive — no dominant AI-first answer-grading player exists yet.
Target User
First 1,000 customers: UPSC GS mains repeaters (second or third attempt) aged 22–28 in cities like Patna, Bhopal, Lucknow, Jaipur, and Chandigarh. These users already spend ₹30,000–80,000/year on test series and printed material, and are active on Telegram study groups. Purchase trigger: failing GS mains with decent prelims scores — a well-defined cohort reachable through Telegram channels and YouTube UPSC creators.
Business Model
₹999/month or ₹7,999/year subscription (₹667/month effective). Targeted at 1.2 lakh addressable "serious repeaters" in Year 1; even 5,000 subscribers = ₹5Cr ARR. CAC via Telegram channel partnerships (₹500–800 per subscriber) gives LTV:CAC > 10× at 12-month payback. Gross margin ~82% (LLM API costs ~₹1.50 per evaluation, ~300 evaluations/user/month = ₹450 COGS). Add a ₹2,999 one-time interview prep module for top-100 scorers.
Competitive Landscape
- Direct (India): Drishti IAS online portal (no AI grading), Unacademy (video-first, no answer evaluation), PaperTyper (generic, not UPSC-specific)
- Direct (global reference): Khanmigo (Khan Academy AI tutor, US) for subject-level feedback; Gaokao AI graders used by provincial education boards in China
- Why we win: UPSC-specific training data (model answers, DOPH guidelines, topper scripts) creates an accuracy moat generic LLMs lack; a flywheel where each graded answer improves calibration
6-Month Plan
- Month 1 (₹1.5L): Fine-tune base LLM on 5,000 annotated UPSC GS answers; build answer-input web UI; launch waitlist via 3 Telegram channels
- Month 2 (₹2L): Beta with 200 aspirants; iterate on feedback format; integrate UPI subscription via Razorpay
- Month 3 (₹2.5L): Public launch; onboard 500 paying subscribers; add Essay paper support
- Month 4 (₹1.5L): Mobile app (React Native wrapper); daily answer challenge with leaderboard
- Month 5 (₹1L): State PSC paper variants (UPPSC, MPSC); affiliate program with UPSC YouTubers
- Month 6 (₹1.5L): Interview prep module; target 2,000 subscribers (~₹20L ARR run-rate)
- Total spend: ₹10L; remaining ₹10L as runway buffer
Risks
- LLM hallucination on factual content (High likelihood × High impact): AI may reward plausible but incorrect content — mitigated by human-in-the-loop spot checks and a user-flagging flow with refund SLA
- Incumbent copy-risk (Medium × High): Unacademy or BYJU's could add AI grading — countered by proprietary topper datasets and UPSC-specific prompt engineering that takes 12+ months to replicate
- Low conversion from free Telegram users (Medium × Medium): Aspirants accustomed to free PDFs may resist paid subscriptions — mitigated by freemium (5 free evaluations/month) and social proof via leaderboard rankings
Score Breakdown
- Market (16/20): 13L+ UPSC + 25L state PSC candidates annually; serious repeaters at ₹999/month yield ₹150–200Cr+ TAM. Not full 20 because it is a test-prep sub-segment, not all-edtech.
- Capital (13/15): MVP needs 2 engineers + LLM API credits + fine-tuning compute — fits ₹8L comfortably; remaining ₹12L is 9+ months runway. No hardware or inventory.
- Team (9/10): 2 full-stack engineers and 1 UPSC content expert (former IAS aspirant-turned-trainer) can ship v1 in 6–7 weeks — near-perfect team-constraint fit.
- Trend (11/15): Backed by TechCrunch-documented investor momentum in AI India niches (Accel/Prosus, March 2026) and HMD AI chatbot signal (May 2026) showing localized AI appetite. No viral Play Store chart signal for exam AI yet.
- Moat (10/15): Data flywheel from annotated UPSC answers and user practice history creates switching costs; network effects limited until community features ship in Month 4.
- Economics (13/15): 82% gross margin, LTV:CAC > 10×, ₹999/month price point validated by existing test-series spend. Slight deduction for uncertain conversion in a free-PDF-accustomed market.
- Speed (8/10): 6–7 weeks to a paying user using existing LLM APIs; fine-tuning can be parallelized with front-end build to hit the timeline.