I built an end-to-end AI calendar agent that passed 11/11 automated tests and scored 3/3 on ground-truth evaluation. AgentEval-Lab evaluates each scheduling output against predefined data, logs agent trajectories, and presents accuracy results in a live Streamlit interface.
On a freelance n8n project, I designed and maintained an autonomous agent workflow that reviewed, classified, and corrected outputs. Prompt-engineered conditional logic and live webhook/HTTP API integrations cut output errors by ~60% over three iterations.
At Jupiter Hi-Tech Systems, I research and score suppliers against structured benchmarking criteria, cutting manual sourcing effort by ~40%. I also designed a scoring rubric that shortened time-to-shortlist from days to hours and translated findings into plain-English reports for senior management.

