Common questions about Pooya's work, AI services, and how to start a project together.
Pooya Golchian is an AI Transformation Lead and AI Product Engineer in Dubai, UAE. He has shipped software since 2009, the last three years in production AI. His work spans agent harness design, MCP servers, coding-agent workflows, RAG and context engineering, and LLM eval suites. He also runs self-hosted open-weight models and ships agentic AI inside Next.js and Node.js stacks. He publishes SpecForge, an open-source method for building software the agentic way.
Pooya Golchian builds production AI agents, the harnesses around them, and full-stack systems in Python, TypeScript, Go or Java. His AI work covers agent systems, MCP servers, RAG and evals. The harness adds scoped tool access, guardrails, and deployment on self-hosted infrastructure under PDPL and DIFC. He works with Claude, GPT, Gemini, Llama and Mistral.
Pooya Golchian builds the harness because it decides whether an agent survives production. The harness scopes tool access through MCP servers and gives the agent a spec to work from, not a prompt to improvise against. An eval suite checks every change against a golden dataset of known-good answers. Cost and rate guardrails apply per role, and your security team can read the audit log. The models are commodity in 2026. Claude, GPT and Gemini all write competent code. Open-weight models like Llama, Mistral and Qwen run on self-hosted hardware when PDPL or DIFC residency rules require it. The harness and the evals separate a working system from a stalled pilot. So do the engineers who own both.
Latest Posts
AI Engineering Blog
Working notes from shipping agentic AI into production. Agent systems, MCP servers, RAG that cites its sources, and evals that catch drift before your customers do. Also the AIDLC method I use to get agent work past the pilot.
