I design and build data and AI systems for engineering and GTM teams at funded startups and enterprises. End-to-end — process audit, agent build with managed agents, MCP, API and auth, validation architecture for compliance-adjacent workflows, adoption coaching, and ongoing maintenance. Direct work, no account managers, no offshore handoff.
How an engagement runs
Three shapes. Most work starts in one and moves to another.
Audit
A short, fixed-scope pass for teams that aren't sure where to start. Process and data review, an architecture recommendation, and a build plan concrete enough to hand to any team — mine or yours. Usually one to two weeks.
Build
End-to-end implementation. I scope it, architect it, build it, and hand it over documented — custom agents, MCP integrations, RAG over internal knowledge, evals, and the interfaces and pipelines around them. Fixed milestones, working software at each one.
Embedded
Fractional architect for a team already shipping. Ongoing — architecture review, the hard parts, evals, and keeping agents healthy in production. I take a few engagements at a time. Mid-project and something's gone sideways is fine too — most of the failure modes are familiar.
Custom models
Also custom post-trained open-source models — Qwen, Kimi — self-hosted on your infrastructure when you need to own the stack, run on private data, or hit cost and latency targets the frontier APIs can't.
What that looks like
- Custom agents and agent orchestration
- Existing app and data integration
- Web scraping, data cleaning, ETL pipelines
- Web interfaces and dashboards
- Agent evals and optimization
- Post-trained open-source models, self-hosted on your infra
- Custom classifiers and task-specific small models
- Document stores and knowledge management