Visitar URL original
starting-ragchatbot-codebase/CLAUDE.md at main · rfencl/starting-ragchatbot-codebase · GitHub
Skip to content

CLAUDE.md

Latest commit

 

History

History
94 lines (60 loc) · 2.53 KB

CLAUDE.md

File metadata and controls

94 lines (60 loc) · 2.53 KB
Raw

CLAUDE.md

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

Running the Application

Start the RAG system:

./run.sh

Or manually:

cd backend && uv run uvicorn app:app --reload --port 8000

Install dependencies:

uv sync

Environment Setup

Create .env file in root with:

ANTHROPIC_API_KEY=sk-ant-api03-H8o...wAAA

Architecture Overview

This is a RAG (Retrieval-Augmented Generation) chatbot system that answers questions about course materials using semantic search and Claude AI.

Core Components

RAG System Flow (backend/rag_system.py)

  • Main orchestrator that coordinates all components
  • Handles document ingestion, query processing, and response generation
  • Uses tool-based approach where AI can call search functions

Vector Storage (backend/vector_store.py)

  • ChromaDB-based semantic search with two collections:
    • course_catalog: Course metadata (titles, instructors, lessons)
    • course_content: Actual chunked course material
  • Supports filtering by course name and lesson number

Document Processing (backend/document_processor.py)

  • Converts course documents into structured Course/Lesson objects
  • Chunks content into 800-character segments with 100-character overlap
  • Supports PDF, DOCX, TXT formats

AI Generation (backend/ai_generator.py)

  • Uses Anthropic Claude (Sonnet 4) with tool calling capabilities
  • Maintains conversation history via session management
  • Can invoke search tools during response generation

Search Tools (backend/search_tools.py)

  • Implements CourseSearchTool that AI can call
  • Handles course name resolution and content filtering
  • Returns formatted search results with sources

API Structure

FastAPI application serving:

  • POST /api/query: Main chat endpoint
  • GET /api/courses: Course analytics
  • Static frontend files at root /

Key Configuration

All settings in backend/config.py:

  • Chunk size: 800 chars, overlap: 100 chars
  • Max search results: 5
  • Conversation history: 2 exchanges
  • ChromaDB path: ./chroma_db
  • Embedding model: all-MiniLM-L6-v2

Document Loading

On startup, automatically loads all documents from ../docs/ folder. Avoids reprocessing existing courses by checking titles in vector store.

Session Management

Tracks conversation history per session for context-aware responses. Sessions created automatically if not provided in requests.