Saad A.

I build AI agents that ship into real products — not demos. LangGraph and RAG systems on the inside, Next.js and Node on the outside, deployed and maintained in production. Latest: an agentic operating system running a robotic biotech lab (Draper-backed, $1M seed), and AI job-matching agents live on a multi-region job board (six contracts, repeat client). Top Rated, 100% Job Success, every client rating 5.0. WHAT I BUILD → Agentic systems LangGraph state machines with checkpointed threads, human-in-the-loop approval gates, tool use, and real-time streaming — exposed as APIs your product calls. OpenAI, Claude, and Gemini models; LangChain where it fits. → RAG and retrieval Embeddings, reranking, and vector search (Qdrant, ChromaDB) over your documents. Open-source models served on RunPod, vLLM, and Ollama — including custom embedding and reranker models (Nomic) when hosted APIs aren't the right fit. → AI workflow automation n8n, GoHighLevel, and custom Python: lead routing, follow-ups, data sync, scheduled posting bots, CRM automation — with an agent in the loop where judgment is needed. → The product around the AI Next.js/TypeScript frontends; Node (Express, NestJS) or FastAPI backends; PostgreSQL, MongoDB; deployed on AWS, Google Cloud, or a VPS you own. SELECTED WORK • Agentic OS for a robotic biotech lab — one conversational interface (Next.js + FastAPI) orchestrating 5 microservices and 6 physical subsystems in an incubator-integrated workcell. LangGraph state machine, Postgres-checkpointed threads, HITL approval gates, SSE streaming; protocol engine covering 8 cell-culture operation types, batching up to 40 flasks with pre-execution validation. Demoed live at an industry conference. • AI job-matching agents for a multi-region job board — agents that match applicants to jobs and suggest roles from a user's profile and CV; migrated the platform off a failed WordPress build onto a custom React/Node stack on a VPS. Six contracts, all rated 5.0. • No-code AI agent builder (SaaS on n8n) — RAG pipeline with Nomic embeddings and a reranker on RunPod, multi-document summarization agents, a Polish LLM (PLLuM) served through vLLM; agent setup time cut from hours to minutes. • Social media content agent — a multi-step agent that generates posts modeled on high-performing content, deployed on a server and scheduled through n8n with Google Sheets and social APIs. HOW I WORK • Architecture first. I'll tell you which parts should be an agent, which should be a plain function, and why. • Production from day one: state, approvals, and failure handling are designed in, not bolted on. Typed code, Git workflow with review. • Clear, structured communication and end-to-end ownership from architecture to deployment. Clients consistently rate me collaborative and a clear communicator. STACK AI: LangGraph, LangChain, OpenAI / Anthropic / Gemini APIs, RAG, embeddings & reranking, Qdrant, ChromaDB, vLLM, Ollama, RunPod Automation: n8n, GoHighLevel, Python Web: Next.js, React, TypeScript, Node.js, Express, NestJS, FastAPI, Tailwind CSS, PostgreSQL, MongoDB, MySQL, Prisma Infra: AWS (EC2), Google Cloud, VPS (AlmaLinux), Heroku Open source: memvault — persistent long-term memory for AI agents. Microsoft Azure AI certified. If you have a product and want agents inside it that survive real users, send me the problem and the stack — I'll reply with an approach, not a pitch.