At Blaze, I helped build the AI CFO, an agent that handles financial operations for businesses, from checking balances to sending cross-border transfers.

Much of my work focused on the TypeScript backend that let agents discover and use paid model and data APIs through x402, with payments settled in USDC. I worked on routing requests to providers and handling payments, with spending limits to keep agent costs in check.

I also built the usage accounting layer so we could trace each API call back to the agent that made it and understand its cost and margin. The payment infrastructure included integrations with Anthropic and Coinbase CDP, backed by more than 125 tests.

At the Tsai Lab, I built machine learning systems to study patterns in social behavior as part of autism research. My work turned data from controlled social interaction experiments into measurable features that could be used for analysis.

I also built a pipeline to automate annotation and feature extraction, reducing preprocessing time by more than 76%. This made it faster to prepare the experimental data for model training and further analysis.

At Dell, I built backend services in Go and Python and worked on how they processed events using Kafka and MQTT. I also containerized services with Docker and Kubernetes, bringing deployment cycles down from hours to minutes.

Another part of my work was making those services easier to operate. I set up secrets management, monitoring, and distributed tracing across more than 10 services so we could follow requests through the system and investigate issues.

At Manochetana, I worked on the website and digital tools supporting accessibility-focused fundraising and outreach in the United States and India. I redesigned the donor onboarding flow, increasing donation conversion rates by 3.4x.

I also built tools for targeted outreach and simplified recurring fundraising tasks. Alongside that work, I helped grow monthly website traffic from roughly 800 to more than 12,500 visitors.

At Nova, UT Dallas's autonomous driving project, I worked on the data and perception systems behind the vehicle. I built a pipeline that processed more than 1.2 million frames of LiDAR, camera, and GPS data each week, and worked on reducing model training time.

I also built a model that combined LiDAR and camera data to assess route hazards. It improved hazard detection precision while reducing false positives by 52%.