MyAcademicCopilot 🎓🤖

An AI-powered study assistant for academic courses

MyAcademicCopilot is a full-stack AI assistant that helps students understand course material. It indexes course PDFs/notes and lets the user ask natural-language questions, using a RAG (Retrieval-Augmented Generation) pipeline and AI agents.

---

🔍 What it can do

• 📂 Ingest course material – PDFs and text files (lectures, notes, solved exercises)
• ❓ Answer questions in natural language about the uploaded material
• 🧠 Use RAG – retrieve the most relevant chunks before calling the LLM
• 🧩 Agentic behavior – different “modes” (answer / summarize / explain) via LangGraph
• 📑 Cited answers – return both the answer and the supporting document passages
• 📊 Observability – log all LLM calls with Langfuse for debugging and evaluation
• 💬 Chat-style UI – React/Vite frontend that feels like a modern AI chat app

---

🏗️ Architecture & Project Structure

The project is split into two main parts:

MyAcademicCopilot/
  backend/          FastAPI server, RAG & agents
    app/
      agents/       Agent definitions and routing
      LLM/          LLM client abstraction (LiteLLM )
      rag/          RAG pipeline: loader, splitter, retriever, store
      schemas.py    Pydantic request/response models
      main.py       FastAPI app entrypoint
    data/
      *.pdf         Example course materials
      intro_ai.txt  Example text knowledge source
    requirements.txt

  frontend/         React + Vite chat UI
    src/
      components/   Chat UI components (header, sidebar, messages, input, etc.)
      App.jsx       Main app component
      api.js        Client for talking to the backend

🔄 High-Level Flow

1. Document Ingestion

   • User provides course materials (PDFs or text files).
   • Backend loads and parses the documents.
   • RAG component splits text into chunks and stores embeddings.

2. User Question

   • User sends a natural-language query via the chat interface.
   • Frontend sends the request to the FastAPI backend.

3. RAG Retrieval

   • Relevant document chunks are retrieved from the vector store.
   • Retrieved context is attached to the LLM prompt.

4. AI Agent Execution

   • LangGraph selects the appropriate agent (answer, summary, explanation, etc.).
   • The LLM (via LiteLLM) generates a grounded answer.

5. Response Construction

   • Backend returns:
     • The answer
     • The supporting citations (document passages used)
     • Metadata for debugging/observability (Langfuse trace IDs)

6. Frontend Display

   • The React UI renders the conversation in a chat format.
   • Citations and context are shown alongside the assistant’s response.

🧰 Tech Stack

Backend

• Python, FastAPI
• LangChain, LangGraph
• **LiteLLM **
• ChromaDB (in-memory vector store)
• Pydantic & Pydantic-Settings
• Langfuse for observability + tracing
• Uvicorn ASGI server

Frontend

• React
• Vite

General

• REST API communication (JSON)
• Modular, clean architecture
• Docker support for easy deployment (full-stack)

🚀 Run MyAcademicCopilot on Your Machine (Docker)

# 1. Clone the repository
git clone https://github.com/bengarusi/MyAcademicCopilot.git
cd MyAcademicCopilot

# 2. Set up the backend environment file
cd backend
cp env.example .env       # Windows PowerShell: copy env.example .env
# Edit the .env file and insert your OPENAI_API_KEY
cd ..

# 3. Build and run the entire system (backend + frontend)
docker compose up --build

# 4. Access the application
# Frontend:  http://localhost:5173
# Backend:   http://localhost:8000

# 5. Stop the system
docker compose down