1
Agentic Hybrid RAG Research Assistant
🚀 After nearly 2 months of building, experimenting, and refining, I’m excited to finally share my Agentic Hybrid RAG Research Assistant!
🔗 Try it: https://pdf-rag-frontend-dmwpm2sv9-prana-labs-projects.vercel.app/
connect me on LinkedIn - https://www.linkedin.com/in/priyanshu-rana-13a678222/
I wanted to go beyond the typical “upload a PDF and ask questions” chatbot, so I built a system that can intelligently work across multiple documents and choose the right retrieval strategy based on the query.
The goal was to make the research experience more accurate, flexible, and verifiable — while giving the system the ability to decide how each query should be handled.
🧠 Query-Aware Agent Routing
The system classifies each query and routes it through the appropriate path:
• Casual conversation → Direct LLM response
• Conceptual questions → RAG Dense semantic search using FAISS
• Technical/keyword-heavy queries → Hybrid RAG search using FAISS + BM25 + Reciprocal Rank Fusion (RRF)
📄 Multi-PDF Research
Upload multiple PDFs and ask questions across all of them, while ensuring relevant information from different documents is considered.
📌 Source Citations
Every generated answer includes the relevant document name and page number, making it easier to verify the information and trace the answer back to the source.
⚙️ Tech Stack & Engineering
• FastAPI + LangGraph for stateful, multi-turn conversations
• FAISS + BM25 + RRF for hybrid retrieval
• Redis for conversation state management
• Groq (Llama 3.1) for fast LLM inference
• React + Vite for the frontend
• Structured output schemas to generate consistent learning-point answers
• LangSmith for tracing, latency monitoring, and observability
💡 What I learned
Working on this for the last two months gave me hands-on experience with
Orchestration, RAG architecture, query routing, hybrid retrieval, A1gent workflows, State management, Structured outputs, and Observability.
There were plenty of things to experiment with, break, debug, and improve along the way — and that’s what made this project such a valuable learning experience.
This is a step beyond building a basic “chat with your PDF” application, and I’m excited to keep improving it.
💻 GitHub: https://github.com/Prana-labs/Agent.git
I’d love to hear your feedback and suggestions! 🙌
Add a comment
0/2000