← Om Badgujar
Project / 01
Agentic AI · RAG
Enterprise Knowledge Intelligence Platform
An Agentic AI knowledge assistant for enterprise document Q&A. A LangGraph intent classifier routes requests between knowledge search, general chat, and tool execution — with Gmail-integrated actions and a full RAG pipeline underneath.
§ Problem
Why it exists.
Enterprise teams sit on piles of PDFs and need answers, not links. A vanilla RAG bot answers questions but can't act; a chat bot can act but hallucinates on internal docs. The goal was one assistant that decides which mode to use per query.
§ Approach
How it works.
- 01Built a RAG pipeline: PDF parsing, chunking, Hugging Face embeddings, ChromaDB vector store, and semantic retrieval.
- 02Designed an LLM-powered LangGraph intent classifier to route each query to Knowledge Search, General Chat, or Tool Execution.
- 03Wired LangChain tools including the Gmail API for AI-drafted, auto-sent emails.
- 04Exposed the system as FastAPI REST endpoints, a Streamlit frontend, and packaged it with Docker + Docker Compose.
§ Key features
What it does.
- LangGraph intent router picks Knowledge Search, General Chat, or Tool Execution per query.
- RAG pipeline with HuggingFace embeddings + ChromaDB and inline source citations.
- Gmail tool for AI-drafted, auto-sent emails from natural-language commands.
- FastAPI REST API + Streamlit UI, containerized with Docker Compose.
§ Tech stack
Built with.
FastAPILangChainLangGraphGroq LLMChromaDBHugging FaceDockerStreamlit
- ◆Single assistant handles document Q&A, chit-chat, and real actions like sending email.
- ◆Grounded answers with retrieval citations from the source documents.
- ◆Containerized deployment — clone, compose up, done.