Algobench

From years of competing in Google Hash Code, I noticed that many industry optimization problems follow the same pattern: messy real-world constraints, a clear scoring function, and the challenge of finding the best solution you can. Most real-world optimization is, at its core, a Hash Code problem.

Algobench was born from this insight. It is a framework to automatically generate and evaluate heuristics for optimization problems. I believe the standard manual tweaking for improved solution quality can be massively accelerated. With AI, you can auto-generate improved algorithms because the problem is perfectly verifiable.

From my experience, the hard part isn't algorithm tuning (that's actually fun, you can make hackathons out of it). The hard part is boxing real-world mess into a well-defined optimization problem: What's feasible? What does solution quality even mean? Once you have that, the solution search can be automated.