Agentic AI Systems Developer — Karachi, Pakistan

I build production-shaped agentic AI systems: routed multi-agent graphs, RAG pipelines with grounded citations, MCP-exposed tools, and the full-stack apps that put them in front of real users.

Core stack: LangGraph · FastAPI · MCP · RAG (ChromaDB + sentence-transformers) · React · PostgreSQL

LangGraph router · MCP tools · RAG · streaming SSE · Repo

Production-shaped support agent. A flash-lite router classifies each turn into rag / tool / escalate before any expensive work runs. Tools (order lookup, product info, ticket creator) are exposed both in-process and over MCP (FastMCP) for protocol correctness. End-to-end token streaming via LangGraph astream_events v2 → FastAPI SSE → React ReadableStream. Auto-escalation creates TKT-YYYY-NNNN tickets in Postgres.

Stack: FastAPI · LangGraph · MCP · ChromaDB · PostgreSQL · React/Vite · Docker · Gemini 2.5

Next.js 16 · AI receipt scanning · approval flows · Live · Repo

Built to solve a real problem in my own shared flat. Multi-tenant houses with admin approval flows for expenses, contribution pools, rent/WiFi/water tracking, and AI-powered receipt scanning that auto-fills expense data. Security-hardened: rate limiting, IDOR protection, CSP, bcrypt + JWT in httpOnly cookies.

Stack: Next.js 16 · React 19 · TypeScript · Prisma · PostgreSQL · Upstash Redis · Claude API · Tesseract.js · Web Push

Cited document Q&A over user-uploaded PDFs

Retrieval-augmented Q&A pipeline with MMR retrieval for diversity and source-cited responses, so users see exactly which chunks the model leaned on.

Stack: Python · LangChain · ChromaDB · HuggingFace all-MiniLM-L6-v2 · Claude API · FastAPI

• Multi-agent coordination patterns and shared memory
• A2A (Agent-to-Agent) Protocol
• LLM observability and production tracing (LangSmith)

📧 sindhikhalid126@gmail.com · 💼 LinkedIn · 🐦 @KhalidUnar27322

Open to freelance and remote roles in agentic AI and full-stack development.