Skip to content

CLAUDE.md

Latest commit

 

History

History
72 lines (49 loc) · 2.33 KB

CLAUDE.md

File metadata and controls

72 lines (49 loc) · 2.33 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

This is a Course Materials RAG (Retrieval-Augmented Generation) chatbot - a full-stack web application that answers questions about course materials using semantic search and Claude AI.

Commands

Always use UV to run the server and manage dependencies, never pip.

# Install dependencies
uv sync

# Run the application (starts FastAPI server on port 8000)
./run.sh
# Or manually:
cd backend && uv run uvicorn app:app --reload --port 8000

# Web interface: http://localhost:8000
# API docs: http://localhost:8000/docs

Environment Setup

Copy .env.example to .env and set ANTHROPIC_API_KEY.

Architecture

Query Flow

Frontend (script.js)
    → POST /api/query
    → app.py endpoint
    → rag_system.query()
    → ai_generator.generate_response() [calls Claude API with tools]
    → Claude decides whether to use search_course_content tool
    → If tool used: search_tools.py → vector_store.py → ChromaDB
    → Response + sources returned to frontend

Key Components

  • rag_system.py: Central orchestrator that coordinates all subsystems
  • ai_generator.py: Claude API integration with tool-use support. Claude decides when to search (tool_choice: auto)
  • search_tools.py: Tool definitions for Claude's function calling. CourseSearchTool wraps vector store searches
  • vector_store.py: ChromaDB wrapper with two collections:
    • course_catalog: Course metadata for semantic course name resolution
    • course_content: Document chunks for content search
  • document_processor.py: Parses course documents, identifies lessons, chunks text with sentence-aware splitting
  • session_manager.py: Conversation history management per session

Configuration (backend/config.py)

Key settings: CHUNK_SIZE=800, CHUNK_OVERLAP=100, MAX_RESULTS=5, MAX_HISTORY=2

Data Models (backend/models.py)

  • Course: Title, link, instructor, lessons
  • Lesson: Number, title, link
  • CourseChunk: Text chunk with course/lesson metadata

Frontend

Vanilla HTML/CSS/JS in frontend/. Uses marked.js for markdown rendering. No build step required.

Course Documents

Sample documents in docs/ are auto-loaded on server startup by app.py lifespan handler.