AI Agent Backend Engineer building production LLM systems
with TypeScript, LangGraph.js, RAG pipelines & multi-agent orc

github

Shubham Bhattacharya  • Peerlist

A fully responsive GitHub homepage clone built using React and Tailwind CSS, focused on clean component architecture, layout structure, and scalable UI composition. The goal of this project was to practice building production-style frontends with reusable components, predictable layout behavior, and responsive design patterns — not just visual cloning.

GitHub UI Clone (React + Tailwind)

Gigi AI Ops — Enterprise AI Customer Support Agent 🤖 Production-grade AI agent backend for D2C e-commerce. Built with zero-trust architecture — every LLM output validated before touching the database. 🔧 Built: - Zero-trust session security — blocks prompt injection & credential spoofing attacks - Hybrid RAG pipeline: BM25 + dense vector search via pgvector for zero-hallucination FAQ answers - Multi-provider LLM fallback: Groq → OpenRouter → Gemini - Human-in-the-loop approval flow for refunds & financial operations above threshold - LLMOps telemetry with cost tracking in USD/INR via tiktoken 🛠️ TypeScript · LangGraph.js · Node.js · PostgreSQL · pgvector · Redis · BullMQ · Zod · Langfuse 🔗 github.com/shubhambhattacharya-dev/Agentic_Project

https://agentic-project-drab.vercel.app/

DocNow — Telemedicine SaaS Platform 🩺 Full-stack platform where patients book verified doctors and consult via live video in under 2 minutes. 🔧 Built: - Live video calls via Vonage WebRTC (no downloads) - Credit-based payments with full audit trail - Role dashboards for Patient, Doctor & Admin - Doctor payout system with admin approval flow 🛠️ Next.js · PostgreSQL · Prisma · Clerk · Vonage · Zod 🔗 docnow-zeta.vercel.app

https://docnow-zeta.vercel.app/

Medo Copilot — AI Launch Readiness Auditor 🚀 Tells you if your app is ready to sell, not just ready to ship. Bridges the gap between "it works" and "it converts." 🔧 Built: - Hybrid scoring: 60% AI insights + 40% Lighthouse metrics - Multi-provider LLM waterfall: Gemini → Groq → OpenRouter - Visual intelligence via Playwright (desktop + mobile screenshots) - Static security scanner for hardcoded secrets & code quality - BYOK support — users plug in their own API keys 🛠️ Next.js · TypeScript · Vercel AI SDK · Playwright · Clerk · Tailwind 🔗 medo-copilot.shubhambhattacharya.dev

https://medo-copilot.shubhambhattacharya.dev/

Built a real-time chat application using Node.js, Express, Socket.io, and MongoDB, focused on scalable event-driven communication and secure authentication workflows. The system supports real-time messaging, user presence tracking, and REST-based authentication, closely reflecting how real-world chat backends are structured. Key Features: Real-time messaging using Socket.io and WebSockets Event-driven architecture for message delivery and presence updates JWT-based authentication with secure login and registration flows REST APIs for user management and authentication Online/offline user presence tracking Cloudinary integration for user profile image uploads MongoDB schemas for users and messages with proper relationships Frontend state management using Zustand Fully responsive UI built with React and Vite Backend deployed on Render APIs tested and validated using Postman Live Demo: https://justchat-d566.onrender.com GitHub Repository: https://github.com/shubhambhattacharya-dev/chatApp Tech Stack: Node.js, Express.js, MongoDB, React, Vite, Socket.io, JWT, Cloudinary, Git, Postman

Shubham Bhattacharya

AI Agent Backend Engineer building production LLM systems with TypeScript, LangGraph.js, RAG pipelines & multi-agent orc