Featured Project

This project delivers an AI-powered search and chat system by integrating advanced storage, retrieval, and AI generation techniques to optimize performance and user experience. It leverages MongoDB Atlas for vector search, keyword search, and hybrid search, ensuring fast and precise information retrieval. The backend is deployed using CloudRun and AWS Lambda, enabling file uploads and context retrieval through CI/CD pipelines and Infrastructure as Code (IaC) automation. Google Gemini API powers conversational AI, while a Pure Chat Mode allows users to disable PDF reading for a focused chat experience. The system improves retrieval accuracy using Jina AI for text embedding and reranking, alongside UI optimizations for seamless cross-device performance. The backend architecture follows Go Backend Clean Architecture, utilizing PostgreSQL for event storage, MongoDB for vector similarity searches, and gRPC for high-performance communication. Key observability tools like Jaeger (tracing), Prometheus (monitoring), Grafana (visualization), and Kibana (log analysis) enhance system reliability. NATS messaging system facilitates event-driven communication, while Elasticsearch improves search indexing and query handling. The frontend, built with Next.js, supports server-side rendering (SSR) and static site generation (SSG), integrating Stripe for payments, Clerk for authentication, and AI SDK for chat functionality, and is deployed using Vercel. The Retrieval-Augmented Generation (RAG) pipeline enhances AI-driven responses by merging search results with LLM-generated outputs. Since Google Gemini lacks system role adjustments, prompt engineering dynamically reformats user queries, ensuring concise and context-aware responses. The system follows an end-to-end data flow where PDFs are ingested, chunked, embedded, and stored for retrieval, while user queries are embedded, ranked, and displayed in real-time, making this scalable AI solution a powerful tool for search and chat-based interactions.