Vidya Marg

# Vidya Marg Vidya Marg is an AI-powered adaptive learning-path generator. Its goal is to solve a common learning problem: a learner often knows some prerequisites already, does not know which concepts are missing, and therefore ends up following a rigid curriculum that is inefficient for their actual level. Instead of presenting a fixed course, Vidya Marg constructs a prerequisite knowledge graph for a target skill, assesses the learner's baseline understanding through an adaptive quiz, and then derives a personalized learning path from the concepts that still need to be learned. The project combines LLM generation with graph theory. The frontend provides the interactive visual experience, while a Python FastAPI backend handles graph generation, assessment logic, path construction, external resource discovery, and persistence. ## Technologies ### Next.js Provides the web application framework and routing for the learning experience. ### React Builds the interactive learning interface. ### Tailwind CSS Provides the styling system. ### Framer Motion Adds animated interactions and transitions to the learning experience. ### React Flow Visualizes the generated prerequisite graph so users can understand how concepts depend on one another. ### FastAPI Provides the backend API and routes for graph generation, quizzes, learning paths, and health checks. ### Python Implements the backend domain logic and learning algorithms. ### Groq Provides the LLM integration for generating prerequisite concepts and adaptive assessment content. ### NetworkX Represents and validates the prerequisite graph as a directed acyclic graph and supports graph traversal used by the learning-path engine. ### Supabase / PostgreSQL Persists sessions, graphs, quiz state, quiz results, paths, and cached learning resources. ### BeautifulSoup Used by the path engine to scrape and cache useful external learning resources returned by search. ### Pydantic Provides typed request/response models and backend data validation. ## Architecture ```text Learner │ ▼ ┌───────────────────┐ │ Next.js │ │ React + UI │ │ React Flow │ └─────────┬─────────┘ │ HTTP ▼ ┌───────────────────┐ │ FastAPI API │ ├───────────────────┤ │ Graph Router │ │ Quiz Router │ │ Path Router │ └───────┬───────────┘ │ ┌──────────────┼───────────────┐ │ │ │ ▼ ▼ ▼ ┌────────┐ ┌──────────┐ ┌────────────┐ │ Groq │ │ NetworkX │ │ Supabase │ │ LLM │ │ DAG Logic│ │ PostgreSQL │ └────────┘ └──────────┘ └────────────┘ │ ┌─────▼──────┐ │ Learning │ │ Resources │ └────────────┘ ``` ## Learning Pipeline ### 1. Skill input The learner provides a target skill. ### 2. Knowledge graph generation The backend asks the LLM to identify prerequisite concepts and relationships between them. ### 3. Graph validation NetworkX validates and operates on the generated graph as a DAG. ### 4. Visualization The graph is returned to the frontend and rendered through React Flow. ### 5. Adaptive assessment The quiz engine begins with concepts that can be assessed independently and generates targeted questions through the LLM. ### 6. Knowledge-state update Answers update the session's known and unknown concept sets. ### 7. Curriculum construction The path engine removes concepts the learner already knows and derives an ordered path through the remaining prerequisite gaps. ### 8. Resource discovery The system finds and caches useful learning resources for the concepts in the resulting path. ## Data Model The application uses separate persisted structures for different parts of a learning session: - `sessions` — target skill and session-level information. - `graphs` — graph nodes and edges. - `quiz_state` — concepts assessed, known, and unknown. - `quiz_results` — answer history. - `paths` — generated curriculum steps. - `concept_resources` — cached study resources. ## API Surface ### Graph ```text POST /api/graph GET /api/graph/{session_id} ``` ### Quiz ```text POST /api/quiz/start POST /api/quiz/answer ``` ### Path ```text POST /api/path GET /api/path/{session_id} ``` ### Health ```text GET / ``` ## Local Development ### Backend ```bash cd backend pip install -r requirements.txt ``` Configure environment variables such as: ```env GROQ_API_KEY=... SUPABASE_URL=... SUPABASE_ANON_KEY=... ``` Run the API: ```bash uvicorn main:app --reload --host 0.0.0.0 --port 8000 ``` ### Frontend ```bash cd frontend npm install ``` Create `.env.local`: ```env NEXT_PUBLIC_API_URL=http://localhost:8000 ``` Start Next.js: ```bash npm run dev ``` ## Why the project is interesting Vidya Marg explores a different use of generative AI: the model is not simply generating educational text. It is helping construct a structured representation of prerequisite knowledge, which is then processed algorithmically. That combination is the core idea: ```text LLM ↓ Knowledge Graph ↓ Graph Algorithms ↓ Adaptive Assessment ↓ Knowledge Gaps ↓ Personalized Curriculum ``` The result is an attempt to make a learning path respond to the learner rather than forcing every learner through the same sequence.