VietStoryLens

An AI-powered system that identifies Vietnamese historical landmarks and generates historically grounded narratives using a sophisticated Agentic RAG architecture.

Built with LangGraph using a Router-Verifier pattern for intelligent query routing. The retrieval layer combines CLIP embeddings (FAISS) with GPS-based Haversine re-ranking, reducing identification errors by 30%.

Includes a custom MCP-server (Model Context Protocol) exposing Google Places API and internal Historical Knowledge Bases as tools for the LLM agent. Strict Guardrails and citation logic ensure every generated "story" is grounded in landmarks.json, eliminating hallucination in historical dates and facts.

Tech Stack: Python, FastAPI, LangGraph, FAISS, OpenAI, Flutter, MCP.