fedy-benhassouna/tunisia-hotel-advisor-ai

# Tunisia Hotel AI Advisor This project is designed for a Tunisian travel agency, enabling users to query hotel information in Tunisia and receive accurate, vocalized responses. The backend leverages web scraping, semantic search, and generative AI to deliver up-to-date and personalized hotel information. ## Table of Contents - Overview - Demo - Features - Tech Stack - Request Processing Workflow - Data Initialization Workflow - Setup Instructions - Future Improvements ## Overview The backend powers a voice agent that automates responses to common hotel-related queries, enhancing customer experience by providing 24/7 service. It processes user queries, retrieves relevant hotel data, generates coherent responses, and converts them into audio for a seamless interaction. ## Demo github.com ## Features - **Semantic Search**: Uses Qdrant and FastEmbed for efficient, context-aware hotel information retrieval. - **Generative AI**: Employs Claude Haiku (via Anthropic API) for intelligent response generation. - **Web Scraping**: Leverages Firecrawl to extract hotel data from TripAdvisor. - **Text-to-Speech**: Converts text responses to audio using pyttsx3. - **API Integration**: Exposes a `/ask` endpoint for frontend interaction, returning both text and audio responses. ## Tech Stack - **Language**: Python - **Frameworks & Libraries**: - FastAPI: For building the API. - Firecrawl: For web scraping (TripAdvisor). - Qdrant: Vector database for semantic search. - FastEmbed: For generating text embeddings. - pyttsx3: For text-to-speech conversion. - python-dotenv: For environment variable management. - anthropic: For interacting with Claude Haiku API. - **External APIs**: - Claude Haiku (Anthropic): For response generation. - Firecrawl API: For web scraping. - TripAdvisor: Data source for hotel information. - **Data Formats**: - Embeddings: Numeric vectors for semantic search. - Markdown/HTML: Scraped …