AI features that hold up in production.
Assistants, retrieval, agents and automation, built into your product with evals and guardrails from the first day.
Which customers were affected by last night’s failed deploy?
- Searching incident log
- Reading runbooks
- Checking tenant list
Tenants on the EU cluster saw errors until the automatic rollback finished. No data was lost. Sources: incident log, deploy runbook.
What I build
- AssistantsChat and copilot features inside your product, grounded in your data.
- RAGSearch and answer over your documents, with sources shown.
- AgentsMulti-step tasks that call your APIs, with a human check where it matters.
- Workflow automationRemove the repetitive manual steps your team does every day.
- Evals and guardrailsTest sets, output checks and limits so quality is measured, not hoped for.
Every answer is checked
Retrieval grounds the model; a check step catches bad output before users see it.
How an AI engagement runs
- PickOne use case with a clear owner and a way to measure it.
- PrototypeA working slice on real data, with an evaluation set from day one.
- HardenGuardrails, cost limits, fallbacks and logging.
- ShipRelease behind a flag, watch the evals, then widen.
Common questions
How many clients do you take on?
Three advisory clients at a time, at most. You work with me directly.
How much time do you commit?
Typically 8 to 20 hours a week, agreed in the proposal.
Is there a long contract?
No. Work is scoped up front and retainers can be cancelled any month.
Do you work with pre-revenue startups?
Yes, including teams that have not shipped a product yet.
Which stack do you work in?
Mostly Go, React, MongoDB, AWS and GCP. Advisory work is stack-agnostic.
How do we start?
A free 30-minute call, then a written proposal with a fixed scope or a monthly retainer.
Have an AI feature in mind?
Bring the use case. The first call is free and takes 30 minutes.
Book a Call