Professional work, older projects, experiments, and stronger current projects as they become real enough to show honestly.
Work & projects
Work with some depth
Published professional work and implemented projects, kept close to what I actually did.
RoboticsIn progress
visual study
Research @ EMBER Center (BAIR)Fall 2026–Present
Teaching a ping-pong robot to see for itself
I’m working @ EMBER Center (BAIR) on onboard vision tracking for the humanoid table-tennis robot from HITTER and LATTE-MV, so it can track the ball without relying on a motion-capture room.
The problem
HITTER’s planner sees the ball through nine external OptiTrack cameras and a ball covered in reflective tape. That works in a calibrated lab, but the robot can’t play anywhere else.
Part-time @ Trisetra AI, I develop computer-vision, OCR, preprocessing, and geometric-analysis systems for extracting walls, rooms, labels, and spatial relationships from difficult architectural drawings. I also lead the company website’s Next.js and React implementation.
The problem
Floor plans encode geometry, symbols, text, and relationships in a format that is easy for people to read but difficult for machines to interpret reliably.
Spatial AI
Floor-plan understanding
Python
Computer vision
OCR
Geometry
Next.js
React
Developer toolsCompleted
visual study
Software Engineering Intern @ AMDMay–August 2023
Making error reports easier to understand
As a Software Engineering Intern @ AMD, I designed and deployed a React dashboard for real-time hardware error-log visualization and contributed to a computer-vision pipeline for detecting missing capacitors on GPU boards.
The problem
Make dense hardware error information faster to inspect while improving visual quality checks across more than 10,000 scanned GPU-board images.
Developer tools
Hardware error analysis
React
Computer vision
Convolutional models
Independent work · Scroll →
The build pile
Projects, experiments, and useful little systems—each labeled by what actually exists.
A tested Gymnasium learner environment for comparing how contextual-bandit policies sequence study topics using mastery, recent accuracy, and practice history.
Contextual bandits
Reinforcement learning
Python
NumPy
Gymnasium
Streamlit
Completed
Instagram Relationship Analyzer
A local, privacy-friendly CLI that parses official Instagram JSON and HTML exports, compares relationship sets, and ranks accounts for review using optional interaction weights.
Privacy-first tooling
Data analysis
Python
JSON/HTML
CSV
CLI
Completed
Daily Orbit
A local-first command center for capturing, prioritizing, and tracking technical projects, tasks, and ideas across human and AI collaborators.
Local-first software
Project systems
Next.js
TypeScript
React
In progress
Interactive Technical Portfolio
This site: a responsive personal archive built around technical case studies, real-life material, and custom interactive systems.