Structured Outputs Are Not Correct Outputs
Why valid JSON is only the start, and where deterministic business rules belong in an AI workflow.
Hello, I’m Anton
I build and operate backend systems, and I’m exploring how the same engineering discipline applies to AI software.
From the notebook
Why valid JSON is only the start, and where deterministic business rules belong in an AI workflow.
A backend engineer's look at idempotency, retries, and duplicate side effects in agentic systems.
How familiar testing instincts can shape practical, repeatable evaluation for AI applications.
Things I’m building
AI-assisted job discovery and résumé matching, built as a practical exploration of applied AI.
Engineering focus
Explores a FastAPI boundary for job discovery, résumé data, and AI-assisted matching.
An experiment in AI-assisted trading signals using Jev AI by Typesafe.
Engineering focus
Separates deterministic market indicators from the experimental AI-assisted signal layer.
A planned support backend exploring correctness, idempotency, guardrails, and evaluations for AI systems.
Engineering focus
Planned around deterministic policy validation and idempotent tool execution before AI guardrails and evaluations.
Away from the keyboard
Pickleball, cameras, and whatever I’m learning next.
I’m learning videography and content creation, and I play pickleball whenever I can.
More about me