Skip to content
View ThomasGust's full-sized avatar
:octocat:
Probably coding
:octocat:
Probably coding
  • Triton Robotics
  • Seattle, WA
  • 23:52 (UTC -08:00)

Highlights

  • Pro

Block or report ThomasGust

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ThomasGust/README.md

Thomas Gust

I'm a high school senior at the Seattle Academy of Arts and Sciences (SAAS). Most of what's here is software for underwater robots or machine learning for biology.

Right now: designing a Jetson-based controller for Triton's 2027 vehicle.

Underwater robotics: Triton Robotics

A friend and I started Triton in middle school to compete in the MATE ROV Competition (first Navigator class, now Ranger class). We finished 2nd overall at the 2026 World Championship (9th in 2025 and 11th in 2024). I am CEO, as well as software & electrical lead. All the code relevant to our ROV (Njord) is hosted in three repos:

  • TritonOS: the onboard runtime. Receives pilot commands over ZeroMQ, runs the control loop and thruster mixing, publishes sensor telemetry, and manages the GStreamer camera streams.
  • TritonPilot: the topside control station (PyQt6). Controller input, live video and telemetry, stereo calibration, and logging.
  • TritonAnalysis: mission-analysis tools that run on a separate laptop. Stereo depth and length measurement, photogrammetry, and underwater color correction.

Seattle Channel ran a segment on the team's design process in October 2026. Some code from earlier seasons is in Triton and Triton2024, but these repos are incomplete and mostly experimental.

Computational biology

  • Elastic-ST: a pip-installable Python library for building gene networks from spatial transcriptomics data, using information about each cell's neighbors as well as its own expression. It includes a custom Cython weighted elastic net, so spatial features can be regularized separately from genes if desired. I started it early in my mentorship with Bruker Spatial Biology.
  • Molecumixer: graph neural network pretraining on small molecules for generation and property prediction. I gained experience designing my own training tasks and working with more exotic network architectures.

Earlier machine learning projects

  • BotSigner: sign language recognition from hand landmarks. My model placed highly in Kaggle's Isolated Sign Language Recognition competition, and I built it into a desktop app for an 8th grade capstone project.
  • SoilingNet: models that estimate soiling (dirt, bird poop etc...) and power loss on solar panels from images. I was and still am very interested in AI for climate/sustainability.

Everything else

My account has several older repos and smaller projects. They're a record of how I learned to program, from pygame clones to my first transformers. I also do contract work building games and tools, like TongoGameOfLife for The Paper Crane Factory.

Much of my work from high school isn't hosted online (especially a lot projects with huge datasets or model files). I think my GitHub already has plenty of cool repos, but I certainly wish I had done a better job of keeping track of and organizing my code; I'm currently going through OneDrive (which seems to not have synced many folders), and my old damaged laptop to try to upload some more of this stuff....

Away from the keyboard, I sing bass in my school's jazz choir, play piano and tennis, and co-lead our Ethics Bowl club. I also love learning new things and always try to find some time each week outside of school or on break to sit down and teach myself something new.

Pinned Loading

  1. TritonOS TritonOS Public

    Onboard software for Triton Robotics' underwater ROV (2nd overall, 2026 MATE ROV Worlds): control loop, thrusters, telemetry, video.

    Python 3

  2. TritonPilot TritonPilot Public

    Topside control station for Triton Robotics' ROV: controller input, live multicam video and telemetry, stereo capture, and logging (PyQt6).

    Python 3

  3. TritonAnalysis TritonAnalysis Public

    Mission-analysis tools for Triton Robotics' ROV: stereo depth and length measurement, 3D reconstruction, underwater color correction.

    Python 3

  4. BotSigner BotSigner Public

    Desktop app that recognizes ASL signs from a webcam with MediaPipe landmarks and a transformer. Model placed 72/1,165 on Kaggle.

    Python 8

  5. Molecumixer Molecumixer Public

    Research prototype: multitask self-supervised pretraining of a graph transformer on ChEMBL molecules (PyTorch Geometric, RDKit).

    Python 6 1

  6. Elastic-ST Elastic-ST Public

    A python data science library to build and analyze biological networks from spatial transcriptomics data. Allows inclusion of spatial information into the networks, going a step beyond graphs deriv…

    Python 3