Adaptive learning paths · Concept-level mastery · Socratic AI tutor
Learnflow AI is a production-quality AI learning platform for Grade 12 students preparing for JEE Advanced, covering Organic Chemistry, Inorganic Chemistry, Advanced Physics, and Advanced Mathematics.
It builds a personalised learning path from an initial assessment, teaches with premium lesson material, sharpens weak spots with adaptive quizzes, and offers Ask Socrates — an AI tutor that guides with questions instead of handing over answers.
Sign up → take an initial assessment → Learnflow AI builds your knowledge profile. From there, the dashboard becomes your command center: it surfaces what to study next, what's blocking your progress, and what you've forgotten and need to revisit. You never hand-build your path — the system decides based on where you actually are, not where the syllabus says you should be.
Adaptive Learning Engine — Mastery is tracked per concept using Exponential Moving Averages. Strong performance unlocks harder material; poor performance surfaces easier questions and prerequisite concepts. ≥ 80% moves you forward, < 60% resurfaces the topic with extra practice.
Recursive Remediation — When a concept keeps failing, the prerequisite graph decomposes it into its foundations (e.g. Gauss's Law → Electric Flux → Vector Area → Dot Product) and recommends teaching those first.
Ask Socrates — A floating AI tutor on every lesson page, aware of the subject, chapter, and content on screen. It never reveals the answer — it asks one leading question at a time and guides you to the insight yourself. Powered by NVIDIA NIM with a built-in Socratic fallback if no key is set.
Adaptive Quiz Engine — Supports single correct, multiple correct, integer, and numerical question types with a timer, mark-for-review, and instant evaluation. Powers both the initial assessment and chapter practice.
Spaced Repetition — Concepts you've learned are resurfaced at calibrated intervals before you forget them.
Premium Lesson Library — 31 premium lessons across all four subjects, tightly coupled to the concept graph.
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, TypeScript, TailwindCSS, shadcn/ui, Recharts |
| Backend | FastAPI, SQLAlchemy |
| Database | SQLite (default) · Supabase PostgreSQL via DATABASE_URL |
| Auth | JWT (email + password) |
| AI | NVIDIA NIM (OpenAI-compatible) · built-in Socratic fallback |
The app runs fully offline with zero external setup. Add NVIDIA API keys and/or a Supabase connection string to upgrade — no code changes required.
All live AI calls go through NVIDIA NIM. Models are configured via NVIDIA_MODEL, NVIDIA_MODELS, and NVIDIA_QUESTION_MODELS in backend/.env (see backend/app/config.py for defaults).
| Role | Primary | Fallback chain |
|---|---|---|
| Socrates tutor | stepfun-ai/step-3.5-flash |
stepfun-ai/step-3.5-flash → deepseek-ai/deepseek-v4-flash → deepseek-ai/deepseek-v4-pro → mistralai/mistral-medium-3.5-128b |
| Question generation | stepfun-ai/step-3.5-flash |
Same four-model chain (JSON-stable models only; reasoning endpoints are filtered out) |
Set USE_AI=false to disable all network calls — Socrates uses the built-in fallback and quizzes serve from the seeded question bank.
Install dependencies for both the backend (Python 3.10+) and frontend (Node.js 18+), seed the database, and run both servers. The full setup instructions are in SETUP.md.
The adaptive engine — grading, mastery EMA, question selection and the
prerequisite-aware learning path — is covered by a backend test suite. The tests
run fully offline (AI calls are forced off and an in-memory SQLite database is
used), so no API keys, network or local .db file are needed.
cd backend
python -m venv .venv && . .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements-dev.txt
pytestGitHub Actions (.github/workflows/ci.yml) runs these tests and a frontend
type-check + build on every push and pull request.
MIT — see LICENSE for details.