Mikhail T.

Maybe the person who understood your codebase has left, or maybe nobody ever did. Or the product was built in Cursor or Lovable and no senior engineer has read it yet. I've stepped into that gap for years. I take ownership of the architecture calls, the technical decisions and a documented handover, so you never end up dependent on me either. 25 years in software, cloud and infrastructure. I was CTO for a logistics holding and for an enterprise software vendor, chief engineer on two carrier-grade data centres in Utah and California, and product lead for a US healthcare SaaS that passed Meaningful Use certification. Stanford MBA, eleven years working in the US. On Upwork: $500K+ earned, 10,000+ hours, 4.8 average rating. I took over a business-critical API that was 80% done and carried the last 20% to production: 2,700+ hours across two contracts, both rated 5 stars. My longest engagement is 4,500+ hours on a single product in a full-stack role with an architecture perspective, rated 5 stars. One client called my work "exceptional across AI architecture, full-stack development, and cloud optimization." AI and LLM work. Most products I review today contain AI-written code, and many call an LLM in production. I review both. On the code side: what Cursor, Claude Code or Lovable generated, checked before it reaches paying users, with payment routes, secrets and access first. On the LLM side: prompt and tool design, retrieval quality, evaluation, cost per request, where user data goes, and what the product does when the model is wrong. I introduced AI-assisted development in two client teams with one working rule: every change starts as a written plan, AI writes the code, and a person approves it before it goes live. For agents I work with Amazon Bedrock, the Claude API and OpenClaw, from a single-agent pilot to multi-agent workflows with a human approval step. I define the architecture, the code review standard and the release process, then document them so someone else can run them without me. The first week goes to learning what's really there: what's production-grade, what's held together with tape, and what nobody has looked at in a year. Stack: Node.js or Python on the backend, React or Next.js on the front, Postgres, AWS with Terraform. You can hire me for full ownership or for a few hours a month as a technical adviser. Clients already book single hours with me to ask which AI agent patterns survive production and which ones break. Either way, the first deliverable is the same: an audit of what's actually there, and a 30 and 90-day plan for what to do about it.