Hello
RSS FeedI'm Chaitanya — I test systems that don't behave the same way twice. Flaky browsers, parallel workers, LLM judges, and lately the AI agents built to test other AI.
This is where I write up the parts worth keeping: what actually broke, what the number turned out to be, and what I'd do differently.
Featured
Our AI candidate started as a for-loop
Three turns of chat completion in a for-loop, and a test that asserted six messages existed. Everything after that was the product refusing to stay that simple.
Recent Posts
I gave my browser tests a model that answers in probabilities
Jev answers in 1.2 seconds and costs almost nothing, so I wired it into a Playwright agent loop and measured it against a large model. Almost every number I got back was a property of my own harness.