At Ethara AI, I build coding benchmark tasks from real repository changes and evaluation harnesses that automatically run and score them. I authored tasks with an average model score of 0.57, and no task was fully solved by the evaluated frontier model.
As quality lead on a multi-stage coding-task harness, I set quality gates and the scoring rubric, and make the final call on whether tasks are correct, fair, and ready to ship. I’ve reviewed more than 2,000 AI training tasks and samples, returning about half for rework.
I contribute pull requests and issue reports to WooCommerce. I also built and maintain a production website for Aspiron Systems, and built QueryForge, a natural-language-to-SQL pipeline with schema-aware prompts and SQL validation.
