RAG Chatbot

RAG chatbot agent using Pinecone, openai/ollama for document Q&A.

Overview

This project implements an enterprise-grade RAG (Retrieval-Augmented Generation) chatbot that combines large language models with external knowledge bases. The system processes PDF documents, creates semantic embeddings, and provides accurate, context-aware responses by retrieving relevant information before generating answers.

2 videos demo of the RAG Chatbot system testing.

System Architecture

Core Components:

  1. Cloud Storage: PDF documents storaage in S3 bucket of AWS(cloud storage)
  2. Document Processing: PDF parsing and intelligent text chunking
  3. Embedding Generation: Semantic vector representations using OpenAI/Ollama
  4. Vector Storage: Pinecone cloud database for scalable similarity search
  5. Retrieval System: Top-K similarity search with metadata filtering
  6. LLM Integration: LLaMA 3 or GPT-4 for response generation
  7. Prompt Engineering: Custom prompts for context-aware responses
  8. Interactive UI: easy to use interface for users to interact with Gradio

RAG Pipeline:

The general arquitecture diagram of the proposed RAG chatbot system.

Technical Implementation

Technology Stack:

  • Vector Database: Pinecone (1536-dim vectors, cosine similarity)
  • LLM Options: LLaMA 3 (via Ollama) or GPT-4.1-mini
  • Framework: LangChain for RAG orchestration
  • Document Processing: PyMuPDF for PDF text extraction

Key Features:

  • Multi-document support with cross-document retrieval
  • Conversation memory for context-aware interactions
  • Source citations linking to specific pages/sections
  • Hybrid search with metadata filtering

Deployment

Infrastructure:

  • FastAPI for REST API
  • Docker containerization
  • Kubernetes deployment on AWS EKS
  • Redis caching for performance optimization

Performance Metrics:

  • <3 seconds end-to-end response time
  • 100+ queries per second throughput
  • <2% hallucination rate (thanks to RAG grounding)

Technology Stack

you can checkout :point_right: my github repository for further details, and sorry againt this may be a bit mess and in spanish but it is a demo project used as MVP(minimal viable product).

This project demonstrates expertise in RAG systems, scalable vector databasesPinecone, and production AI deployment(using docker + kubernetes), and monitoring of KPIs(key Performance Indicators). For collaboration opportunities or more clarifications, please contact me.