At Sunsaver, I shipped a Gemini agent for Habit Game that decides whether to hold or return users’ deposits. I took it from in-house testing to a pilot of about 50 customers, then migrated it from OpenAI to Gemini in production using prompt changes.
I also built an end-to-end simulator and two LLM judges to evaluate agent behavior, catch tool failures, and score conversations for future fine-tuning. For Sunsaver’s Solar Platform, I built customer and admin features, Twilio messaging, and a QLoRA fine-tune of a local Qwen model.
At Servo-Robot Inc., I built an object-detection model to fit microcomputer memory and latency limits, then deployed it on-device with existing hardware. My work at Fairfield University includes three peer-reviewed AI research papers, including two as first author, on graph neural networks and language modeling.
