ICAIRA 2026, Istanbul: Autonomous Program Repair with Feedback-Driven Test Synthesis

I presented a solo-authored paper at the International Conference on Artificial Intelligence Research and Applications (ICAIRA 2026) in Istanbul, Türkiye, held 2–4 September 2026. The paper is “Autonomous Program Repair with Feedback-Driven Test Synthesis: An Empirical Evaluation” (Paper ID 193), accepted for oral presentation.

The premise is simple. Automated program repair systems are usually evaluated against a fixed test suite, which quietly rewards patches that satisfy the tests rather than patches that fix the bug. If the suite is the oracle, the suite is also the ceiling. The paper looks at what happens when you let the repair loop synthesise new tests from its own failures instead – using the feedback signal to grow the oracle rather than treating it as fixed – and evaluates that empirically against the standard approach.

Presenting solo across nine time zones at 4:40 in the morning Pacific is its own kind of engineering problem. Worth it: the questions from the room were sharper than the ones I got from the reviewers.