AI agents and LLM features
I was building full-stack systems for more than a decade before AI coding tools existed. Now I use them every day, and I build AI features and agents for clients. That background matters: an LLM feature is still a feature, with an API, a database, costs, failure modes and users. I build it like one.
Who this is for
- Product teams that want useful AI features in an app that already works.
- Businesses with repetitive work that an agent with the right tools could take on.
- Engineering teams adopting AI coding tools that want speed without losing code quality.
What I can do
LLM features in existing apps
Summaries, smart search, drafting, classification, extraction and chat over your own data, added behind your existing API with structured output, sensible fallbacks, cost controls and evaluation.
Custom agents and MCP servers
Agents that use real tools: your APIs, databases, dashboards and documents. I build MCP (Model Context Protocol) servers so agents can use those tools safely, with clear permissions and logs.
Workflow automation
Automations with n8n or custom code that connect AI steps to the systems you already use.
AI-assisted delivery for your team
I use Claude Code, OpenAI Codex, Cursor and Google Antigravity day to day. I can help your team choose and configure them, write project rules and instructions, and set up reviews so AI-written code meets the same bar as human-written code.
How I approach AI work
- Start small. One feature, measured, then expand.
- Keep it swappable. Models change every few months, so the provider sits behind one interface.
- Measure it. A small evaluation set before shipping, so changes are judged on results, not vibes.
- Own the code. Whether a person or an agent wrote a line, I review it and stand behind it.