AI Agent Evaluation & Human Feedback Platform by Bowen JinAI Agent Evaluation & Human Feedback Platform by Bowen Jin

AI Agent Evaluation & Human Feedback Platform

Portfolio project
Bowen Jin

AI Agent Evaluation & Human Feedback Platform

Overview

Designed and implemented agent systems for multi-step reasoning and task execution. The work combined planning, tool use, and feedback loops to support complex workflows while making reliability measurable and improvable.

What I built

Agent workflows that coordinated multi-step planning, tool calls, and feedback loops.
Evaluation frameworks and rubrics for measuring performance, reliability, and failure modes across robotic workflows.
Human-in-the-loop pipelines for collecting, filtering, and incorporating expert feedback into agent behavior.
Data-quality standards, annotation guidelines, and automated checks for high-fidelity training and evaluation data.
Scalable model-serving and evaluation infrastructure, including prompt engineering and iterative refinement of reasoning traces.

Outcomes

Improved task-completion rates by 35% across complex workflows.
Evaluated 50+ distinct task scenarios.
Processed 10,000+ expert annotations each month through human-feedback pipelines.

Focus

The project centered on the practical work behind dependable AI systems: defining quality signals, finding failure modes, tightening feedback loops, and scaling evaluation and deployment infrastructure.

0 likes · 0 views · Nov 21, 2026

Agent systems, evaluation frameworks, and expert-feedback pipelines for reliable AI workflows.

Tags