Hello, It's Trisha

A machine learning engineer based in Houston, TX.
I build AI-powered systems to automate workflows and help catch what breaks behind the scenes.

About

I currently work at Oceaneering International, where I build LLM workflows that extract and surface insights from unstructured data. Previously, I developed machine learning models at the intersection of computer vision and NLP.

Tools and technologies I currently work with:

  • Python
  • Javascript
  • SQL
  • PyTorch
  • PowerBI
  • Azure

When I'm not tinkering with code, you’ll find me strength training, out on a run, or reading anything from historical fiction to books on biology and psychology.

Profile image

Experience

Oceaneering International

Machine Learning & Robotics Engineer

Feb 2025 - Present

  • Reduced report review cycles by ~50% by developing an LLM-driven pipeline on Azure to extract and summarize insights from unstructured documents.
  • Designing monitoring and alerting systems using Azure Application Insights to track system performance, enabling proactive detection of downtime and resource inefficiencies for 20 applications.
  • Built PowerBI dashboards for 3 clients, transforming data into insights to support decision-making.
  • Cut sprint planning overhead by designing automations for team capacity forecasting and performance analysis on JIRA.
  • Delivered $200K+ in R&D savings by analyzing performance data from 25 robot platforms, identifying critical gaps, and informing investment decisions with quantitative insights.

Publications

Text to Graphics by Program Synthesis with Error Correction

Text to Graphics by Program Synthesis with Error Correction

CVPR Generative Models for Computer Vision Workshop (GCV) — 2023

Top Finding: Program synthesis with error correction improves precision and enables reliable procedural rendering in text-to-graphics tasks.

MartiNet: An Efficient Approach For River Segmentation In SAR Images

MartiNet: An Efficient Approach For River Segmentation In SAR Images

IEEE CONECCT, Geoscience and Remote Sensing Technologies — 2022

Key Contribution: Developed a semantic segmentation model that surpassed state-of-the-art performance while using 71% fewer parameters.

Projects

Offscreen

Favorite Reads

Cover of I Contain Multitudes
Cover of The Last Lecture
Cover of Thinking, Fast and Slow
Cover of A Thousand Splendid Suns
Cover of Pachinko