Voice-Powered Multimodal RAG Agent

Agentic RAG with LangGraph and NVIDIA Nemotron for voice-driven multimodal retrieval.

Engineered an agentic RAG system using LangGraph and NVIDIA Nemotron, enabling voice-driven multimodal context retrieval over text and images.

  • Autonomous ReAct-based agentic loop with cross-encoder reranking for retrieval optimization.
  • Integrated Llama-3.1-Nemotron-Safety-Guard for real-time PII detection and content moderation.
  • Deployed on Hugging Face using NVIDIA NIM for scalable, low-latency inference.

Tech Stack

  • Python, LangGraph, NVIDIA NIM, Nemotron, Hugging Face