Rohan Sardar

Rohan Sardar

he/him

ML Student

ChatBot with RAG on PDF

Published:

🔗 Deployed Link: ChatBot with RAG on PDF

Tech Stack

  • Python
  • Streamlit
  • LangChain
  • GROQ
  • LangSmith

Features

Document Ingestion

  • Accepts one or more PDFs as input.
  • Uses LangChain document loaders to split and clean the text.

Embedding & Vector Store

  • Generates embeddings using Hugging Face models.
  • Stores and indexes them using FAISS vector database for efficient retrieval.

RAG Pipeline

  • Retrieves relevant document chunks based on user queries.
  • Feeds the context into an open-source LLM for grounded answer generation.
  • Powered by GROQ for fast and accurate response.

User Interface

  • Built with Streamlit for seamless UX.
  • Users can upload PDFs and ask contextual questions.

Observability & Monitoring

  • Integrated with LangSmith for pipeline debugging and performance tracking.

Workflow

  1. PDF Upload
    Users upload documents; text is extracted and preprocessed.

  2. Embedding & Storage
    Content is chunked and converted into embeddings using Hugging Face models, then stored in FAISS.

  3. Query Handling
    User queries are matched against the vector store to fetch the most relevant chunks.

  4. Answer Generation
    Retrieved context is passed into an LLM, and the generated answer is returned to the user.

Highlights

  • 📄 Multi-PDF ingestion
  • 🔍 RAG-powered contextual answers
  • ⚡ Fast inference using GROQ and FAISS
  • 📈 Observability via LangSmith