Krish Patel

I developed a Go evaluation platform for coding agents across 120 production tasks, reducing merge-readiness disagreement by 57% with deterministic execution, mutation testing, and model adjudication.

I architected distributed Bayesian hyperparameter optimization on Vertex AI, quadrupling trials per run while improving PR-AUC by 8.8% and GMV-weighted PR-AUC by 33.8% for fraud models.

I launched a Google ADK multi-agent platform that synthesized enterprise knowledge into briefs and visuals, cutting content production time by 78% and earning lab-wide adoption.

I created a LangGraph vision-agent system that reconstructed engineering diagrams in AutoCAD with 99.3% symbol detection accuracy, cutting manual work by 85% and saving $350K+ annually.

I helped build the core meal recommendation engine, designed evaluation frameworks to measure recommendation quality, and contributed to additional user-facing product features.