Retrieval-Augmented Generation (RAG) system designed to allow users to interact with the Federal Register dataset via a natural language interface. The system integrates a local LLM (Qwen 0.5B via Ollama), semantic search using FAISS, and a daily-updated MySQL database, enabling accurate and up-to-date responses based on real government publications.
Agentic Chat System: Users can ask natural language questions; the LLM responds using tools (function calls) instead of directly querying APIs
Data Pipeline: Automatically fetches and processes documents from the Federal Register API and stores them in a local MySQL database
Semantic Search: Uses SentenceTransformers + FAISS to find the most relevant documents
Local LLM Inference: Uses Qwen 0.5B model through the Ollama platform
Streamlit UI: User interface to chat with the agent in real time
FAISS – for vector-based retrieval
MySQL – for structured document storage
Ollama + Qwen 0.5B – for local LLM responses
aiohttp + aiomysql – fully async data pipeline
Sentence Transformers – for document embedding
Streamlit – for the chat interface
pip install -r requirements.txt
streamlit run main.py
Fedral Register DB:
RAG - BOT:
