DEVELOPER TOOL / PUBLIC ALPHA
Explore the tool
AI PR Governor
A local CLI for checking repository state and pull-request evidence before acting on AI-generated code. It provides machine-readable decisions and queue planning.
ABOUT THE WORKSHOP
About Adem
I'm Adem, an independent engineer working on AI evaluation, compiler correctness, and developer tools. I also build interactive experiments in physics, graphics, and music.
I enjoy working out why software fails, then building a fix other people can use. My projects usually grow through experiments, tests, and review.
UPSTREAM WORK
Merged upstream contributions
NVIDIA / SKILLEVALUATOR
Evaluation results with more context
Added uncertainty estimates and paired-outcome reporting to help interpret agent-skill evaluations.
Merged upstream · August 2026 ↗ XDSL / COMPILER INFRASTRUCTURESix fixes across compiler transformations
Repaired silent wrong results in loop flattening, floating-point optimizations and dynamic-shape interpretation, alongside loop-rewrite and LLVM-backend checks.
Six merged fixes · September and October 2026 ↗Explore the six xDSL changes
- Preserve iteration counts when flattening loops
- Keep positive and negative zero distinct during optimization
- Preserve special values when folding division by zero
- Use runtime shapes in the linalg interpreter
- Require positive factors when folding loop ranges
- Reject unsupported LLVM call operand bundles
Also contributed a merged test fix to DESC, a stellarator optimization code. See the eight open proposals and current review contributions (checked 8 October 2026).
TOOLS AND EXPERIMENTS
Tools and experiments
LOCAL CREATIVE WORKFLOW
Arkhē Lyric Lock
A local workflow for aligning lyrics to vocals, flagging uncertain timings and repairing them in a browser.
Source and video demos ↗
ARSENAL FIRST DROP
Browser-native physical experiments
Water and cloth experiments built around interaction and documented tests. These are computational demonstrations with stated limits.
Open the demos ↗ · Source ↗WORKING TOGETHER
Working together
I'm interested in practical work on AI evaluation, software reliability and developer workflows. Bring a difficult behavior, an uncertain evaluation or a release decision that deserves better evidence.
My professional record is on GitHub and this site.
I use AI tools in my workflow and document their role in individual projects. Reproducers, tests and review feedback are part of what I share.