Alongside the products, this is where my own investigations and prototypes take shape. What proves itself becomes a product; the methodology behind it carries into every client project.
How do AI agents work reliably instead of just impressively? I run measurement series across hundreds of agent runs and document the findings in a private research paper: test benches, sabotage probes, failure classes, and the rules that follow from them. The rules from it feed into the apps and into client projects.
I'm happy to share the research document on request.Selected work beyond the product line:
Tactics game · Godot
A turn-based co-op game across forty nights. The game engine was ported deterministically from JavaScript to C# against golden test values: every night is exactly reproducible.
Details or a demo are available on request.
Android · Engineering calculator
Heat-treatment data for knifemakers, from curated and referenced sources. It ships only once limit checks and safeguards are in place.
Details or a demo are available on request.Automation
Image generation fully local, with no cloud services. Throughput was measured and tuned from 53 down to 8.5 seconds per image.
Details or a demo are available on request.