Enterprise Knowledge Base with RAG

Role: Lead Architect & Developer Tech Stack: Java 21, Spring AI, PostgreSQL (PgVector), Docker, OpenAI API

📖 Project Overview

This project is an AI-powered Question Answering system designed to retrieve precise information from unstructured enterprise documents. It addresses the specific challenges of context window limits and hallucinations in industrial applications.

🏗️ System Architecture

graph TD
    User[Client] -->|Query| API[Spring Boot API]
    
    subgraph "RAG Pipeline (Spring AI)"
        API -->|Orchestrate| Service[RAG Service]
        Service -->|1. Hybrid Search| Retriever[Retriever Strategy]
        Service -->|2. Augment| Prompt[Prompt Template]
    end
    
    subgraph "Storage Layer"
        Retriever -->|Keyword| ES[Elasticsearch]
        Retriever -->|Semantic| Vector[PostgreSQL / PgVector]
    end
    
    subgraph "LLM Integration"
        Prompt -->|Context+Query| Model[OpenAI / Bedrock]
        Model -->|Response| User
    end

Loading

🚀 Key Engineering Contributions

Hybrid Search Algorithm: Implemented a weighted retrieval strategy combining Elasticsearch (BM25) for keyword precision and PgVector for semantic understanding.

Scalable Pipeline: Architected a microservices-ready ingestion pipeline using Spring AI to orchestrate document chunking and embedding generation.

Context Optimization: Solved context window constraints by optimizing chunk sizes based on semantic boundaries.