◆ ABHIGYAN COGNITIVE LEARNING ENGINE
PRE-SEED · 2026 · INDIA
Abhigyan
A live cognitive engine for India's exam aspirants.
Not another content library, a system that models how each learner thinks, and teaches around it.
Raising ₹2 Cr / $225K PRESS → TO BEGIN
The problem

Test prep in India is static.
Every student walks the same path.

FINISH START CH.1 CH.2 CH.3 CH.4 CH.5 ★
The lucky one · already strong, the path happened to fit. Makes it through.
Overwhelmed · content too fast, too dense. Drops off at the first hurdle.
Wrong pace · already knows this. Bored, skims, retains nothing.
Disengaged · no feedback loop, no challenge. Stops showing up.
Gaps never found · appears to progress, but a hidden weakness is never caught.
The path is the same for all five.
Only one of them needed it.
The insight

Learning isn't a content problem.
It's a diagnosis problem.

Know a learner's cognitive state, their strengths and gaps across the skills each subject demands, and you can orchestrate everything else around it: what to teach, how, and when.

Nobody in Indian test prep models the learner in real time. That's the whole product.
CONTENT-FIRST
One syllabus. The same path for every student.
↓
FOR THE LEARNERHours spent on what they already know, while the real weak spots go unfound.
↓ THE SHIFT
DIAGNOSIS-FIRST
One profile per student. The path adapts to how they think.
↓
FOR THE LEARNEREvery session targets the real weak spot, so progress compounds.
The product

A cognitive profile that orchestrates learning.

SETUP · ONCE onboard baseline PERSONA STATE REFINED EVERY LAP learn evaluate evolve
HOVER / TAP A CAPABILITY

Multi-agent classrooms

Three AI agents plan the lesson, the peer classmates, and the Socratic traps before class even starts.

AI copilot tutor

A tutor that always knows the student's current cognitive profile, not a generic chatbot.

Vision-based evaluation

Photograph a handwritten answer; get structured feedback mapped to specific cognitive skills.

Adaptive interface

Tactical and focused for aspirants, warm for K-12. One engine, two faces.
Every feature is a consequence of one thing: a good model of the learner.
Live simulation

Watch the engine model a learner.

NOW HAPPENING
◷
Sets a goal
Target: UPSC Civil Services. The engine spins up a fresh cognitive profile.
COGNITIVE PROFILELIVE
Recall
–
Reasoning
–
Depth
–
Application
–
0MASTERY
CURRENT FOCUS
Building the profile…
Goal
Baseline
Learn
Evaluate
Adapt
Defensibility

The model is the moat.

PERSONA STATE · UPSC GSLIVE
Polity
82
History
38
Missed 3 of 4 questions on the freedom struggle; could recall dates but not link causes to outcomes. → reasoning gap, not recall.
Economy
64
Geography
45
Strong on physical geography, weak on map-based questions. → diagram/spatial skill below threshold.
01
Proprietary cognitive state

A structured model of how each student thinks, not a prompt anyone can copy.

02
Compounds with use

Accuracy improves with every interaction. A late entrant starts from zero data.

03
Multi-provider routing

Not locked to one model. Margins improve as inference gets cheaper.

04
Built for the constraint

Runs in the browser on any low-end phone, no install, tolerant of patchy internet. Incumbents built for broadband can't follow down-market.

Why Bharat

Built natively for the
Indian exam ecosystem.

PYQs Mains strategy Hinglish Optional subjects Prelims cutoff
  • The actual language and concepts aspirants use, not translated in after the fact.
  • Exam stages and topic maps, built in from the start.
  • DPDP-compliance roadmap and planned Mumbai-region data residency.
EXAM COVERAGE · BUILT IN
UPSC CSEPRELIMS · MAINS · INTERVIEW
SSCCGL · CHSL · MTS
RRBNTPC · GROUP D
BankingIBPS · SBI PO
Stage maps, topic weightings, and prior-year patterns, modelled per exam.
Market

275M learners. We start with one cohort.

0
Competitive-exam aspirants, civil services, railways, banking
0
K-12 students, CBSE, ICSE, State Boards
0
Total Indian learners on static test-prep tools
First wedge: SSC & RRB aspirants studying online
They already spend ₹5K–50K/year on coaching, they study on their phones, and they're desperate for an edge. Reachable through YouTube coaching communities and Telegram groups, channels we can run without a sales army.
First target: 1,000 active aspirants in 6 months.
Unit economics

Margins that don't break, unlike the last generation.

0–95%
gross margin on AI cost, per active student.
Every frontier model that ships is cheaper than the last. Our margins improve every quarter.
ABHIGYAN · AI inference~₹40–140 / student / mo
LAST-GEN · human tutors + sales army100%+ of revenue
Monthly subscription₹350
Gross margin at this price79%
Illustrative · assumes ~₹75 blended inference cost / student / month
Traction

Working prototype. Full lifecycle shipped.

Auth + data layer
Firebase auth, Firestore, the persistence backbone.
Cognitive engine
Persona-state that updates with every interaction.
Multi-agent classroom
3 agents orchestrate each class before it begins.
Vision evaluation
Grades photographed handwritten answers.
Multi-provider routing
Gemini + DeepSeek, switchable, no lock-in.
Android-ready
Capacitor path wired, ready to ship as an Android app when needed.
Observability
Langfuse traces + prompt management in production.

Built solo, end-to-end. The hard parts (orchestration, vision, routing) already work. With no revenue yet, the build depth is the traction.

▶ 90-second demo
The ask
₹2 Crore / $225,000 iSAFE / SAFE · [ equity % ]
PHASE 01
0–6 MONTHS

Launch the SSC / RRB wedge. Reach 1,000 active aspirants. Iterate hard on retention.

PHASE 02
6–12 MONTHS

Expand to UPSC CSE. 5,000+ active. Prove 40%+ monthly retention.

PHASE 03
12–18 MONTHS

Seed-ready. Adjacent exams. A clean unit-economics case at scale.

USE OF FUNDS
40% Eng
30% Inference
20% Content
10% GTM
Senior engineer, orchestration & inference optimization Inference + infra Exam content & data GTM in the wedge
This cheque de-risks the seed by proving the model works at scale.
The team

Built by people who know the problem from the inside.

Pragnan Chakraborty
SENIOR CONVERSATIONAL AI DESIGNER · FELLO AI
M.A. Computational Linguistics, The English and Foreign Languages University. Built Abhigyan end-to-end across multi-agent orchestration, cognitive engine, vision evaluation, and AI routing. Expertise spans NLP, Prompt Engineering, Generative AI, and LLMs. Ex Kore.ai.
pragnan-chakraborty
Sourima Laha
SR. SPEECH & LANGUAGE R&D ENGINEER · CERENCE AI
Expert linguist specialising in psycholinguistics and language acquisition in both children and adults. Industry background spans Cerence AI, Tech Mahindra, and Amazon (4+ years as ML Data Linguist). Ensures every learning interaction in Abhigyan maps to how humans actually acquire language and knowledge.
sourima-laha
Rahul Prajapati
LEAD AI/ML · PROXIMITY WORKS · EX KORE.AI
Lead AI/ML engineer with deep background in LLMs, multi-agent systems, and R&D. Ex AI Engineer at Kore.ai. Active in the global GenAI community with 6,000+ followers. Owns the model layer, Vision RAG pipelines, and the engineering infrastructure.
rahul-prajapati
PRODUCT · PEDAGOGY · ENGINEERING · THREE DOMAINS, ONE SYSTEM
01 / 11 COVER