akaLocal Agent Gateway
A loopback gateway between an OpenAI-compatible client and a local Ollama model, with the whole test rig around it.
- 66 automated tests
- 98.71% coverage, CI threshold 90%
- Gateway API 28/28
- LLM evaluation 33/33
- End-to-end 5/5
Hands-on AI Product & Agentic Systems Architect
I build AI products end to end — from scenario and model to secure remote delivery and verification.
I define safe boundaries for agent actions and prove the result with scenarios, evaluations and automated checks. Every claim below opens in one click.
Saint Petersburg, Russia · UTC+3 · remote · Russian native, English working
Selected work
Four projects, each with the number that backs it. The numbers come from the same data the Russian pages use, so they cannot drift apart.
Psy AI Admin
A site assistant for a psychology centre that answers only from an approved knowledge base — and says so when a live person is needed.
- 8 safety tests
- 38 scenarios
- boundaries fixed before the build, not after
Photo Meta Editor
Batch-edits the capture date of photos and video in the browser. Files never leave the device.
- 19 automated tests
- 0 network requests
- enforced by the page's own CSP, not by a promise
Agent-driven storefront
An assistant takes a shopper from question to a finished cart; three service agents sit behind it. Built with a co-founder, in progress.
- 12 mechanical scenarios
- 8/8 and 4/4 on the current run
- checkable assertions, not model-judges-model
And eleven browser games
Not the main thing, but the proof that things reach real people: they are playable right now, on a phone, with no install. Source is open for all of them.