Company Policy RAG Bot

An autonomous, agentic RAG system built with LangGraph, Pinecone NVIDIA Inference, and Gemini 2.5.

🏗️ System Architecture

🚀 Key Features

• Agentic Logic: Uses LangGraph to dynamically route user queries. The agent intelligently decides when to call the retrieval tool and when to answer directly.
• Server-Side Inference: Leverages NVIDIA Llama-text-embed-v2 hosted directly on Pinecone, bypassing complex client-side embedding math and dimension mismatch issues.
• Persistent Memory: Utilizes MemorySaver to track conversational context across multi-turn user sessions.
• Resilience: Implements programmatic cool-down logic to handle API rate limits (429 errors) gracefully.

🛠️ Tech Stack

• Orchestration: LangGraph
• LLM: Gemini 2.5 Flash
• Vector Infrastructure: Pinecone (Inference-Enabled Index)
• Embedding Model: NVIDIA Llama-text-embed-v2
• Development: Python 3.12+

⚙️ Quick Start

1. Environment Setup

Create a .env file in the root directory with your keys:

GOOGLE_API_KEY=your_gemini_key_here
PINECONE_API_KEY=your_pinecone_key_here