How dmquan1105/travel_chatbot works
Project type and purpose
• This is a full-stack travel chatbot application.
• Frontend: React single-page app for chat UI.
• Backend: Python (FastAPI/uvicorn) exposing REST/WebSocket endpoints.
• Core logic: LangChain + Gemini AI for planning, rewriting, synthesizing responses.
• Data: MongoDB for conversation/message storage, a vector store for travel knowledge, Scrapy crawler for raw travel data, and external APIs (weather, Tavily/web search).
File structure
• Top-level:
• README.md, .env.example, requirements.txt, main.py, prompt.py
• Agents folder: base_agent, orchestrator, planner, rewriter, responder, response_synthesizer, travel_bot
• backend/: server.py (FastAPI app), conversation.py, message.py, database.py
• frontend/: React app (components, assets, config files)
• crawl/: Scrapy project to crawl travel data → produces data/travel_data.jsonl
• travel_vectorstore/: loader.py (ingest), storage.py (persist vector embeddings)
• tools/: wrappers for external APIs (get_weather, web_search, search_travel_info)
• scripts/preload_models.py: preloads vectorstore and models
• utils/: misc scripts for data handling
README for additional insights
• Tech stack: Python 3.8+, Node.js 16+, MongoDB Atlas
• Dependencies: LangChain, Gemini API, WeatherAPI, Tavily API
• Run sequence: crawl → preprocess/load data → start backend → start frontend
• Runtime: frontend on Vite dev server, backend by uvicorn server:app