Pareto / AI evaluation
AI evaluation tools
At Pareto, I contributed to a shared platform for creating AI tasks, running evaluations, and inspecting the results.
The platform
Researchers needed to follow a task from its definition through execution and grading. The interface brought task libraries, runs, and evaluation results together.
My work
I worked on React interfaces, Python APIs, calibration results, grading displays, and execution controls. That included worker routing, cancellation, and investigating failed runs.
One example
A task could meet its target score and still contain a defect. I worked on carrying those findings into the calibration decision so the task could be sent back for revision.