The Work Edit.

At Morgan Stanley, I developed a human-in-the-loop agentic AI system to improve Jira ticket creation and resolution. The system brought together three specialized agents, with human review as part of the workflow.

I built the backend in Python and FastAPI, using LangGraph and MCP for integrations. My work also included evaluating LLM performance and implementing security guardrails, balancing useful automation with oversight and reliable behavior.

The system achieved 75% first-pass accuracy and reduced ticket creation from 30–40 minutes to about 5 minutes—an 83–88% reduction—with adoption by more than 50 developers. Alongside this AI work, I contributed to C++ and Linux development in Sales & Trading Technology.

At the USF RARE Lab, I developed an LLM-driven recipe recommendation experience using a Misty robot. The system combined the Gemini API, a Raspberry Pi, and WebSocket communications to connect recipe generation with a physical, interactive interface.

The experience handled ingredient extraction, dietary preferences, and recipe refinement so recommendations could respond to what a person had available and wanted to cook. I also evaluated the human–robot interaction, examining how clearly the robot communicated and how quickly it responded.

Evaluation recorded 92% ingredient detection accuracy, a 4.6 out of 5 human clarity rating, and interactions under 3 seconds. The work connected language-model capabilities with the practical demands of a responsive robot experience.