Krish Patel

At Coinbase, I built an autonomous coding-agent evaluation platform that measures performance on repository-specific tasks across combinations of substrate, model, agent backbone, and reasoning effort. I created the full task-mining and evaluation pipelines, giving teams visibility into where agents spend money, where costs can be reduced without sacrificing performance, and how to quantitatively improve the substrate.

At Shopify, I built the full distributed hyperparameter optimization system for our fraud models on Vertex AI. It coordinated experiments across multiple workers and shared what each one learned, allowing the team to test four times as many model configurations in the same amount of time and improve fraud detection quality.

At Intact, I created an AI platform that turns internal knowledge into clear briefs and visuals. It reduced the time needed to produce content by 78% and was adopted across the lab.

At Nexxa, I worked with one of Europe's largest rail technology providers to automatically turn large, scanned engineering diagrams into editable AutoCAD drawings. I developed computer vision models to detect symbols and read diagram content, built the agent workflow that reconstructed each drawing, and optimized self-hosted model inference to make the system faster and more memory-efficient.

At PlateHack, I shaped the recommendation engine that turns a person's body metrics, activity, fitness goals, dietary needs, allergies, and food preferences into practical meal plans.