Skip to content
View RubenHaisma's full-sized avatar
7 followers · 5 following

Achievements

Achievement: Pull Sharkx4Achievement: Pair Extraordinairex3Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx4Achievement: Pair Extraordinairex3Achievement: YOLOAchievement: Quickdraw

Highlights

  • Pro

Block or report RubenHaisma

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. Markdown is supported. 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
RubenHaisma/README.md

Hi, I'm Ruben

I build production AI agents at Laava. In the open, I work on computer vision and medical-imaging tooling, mostly on evaluation: metrics that hold up, benchmarks that can be reproduced, and checks that show what a model actually learned.

Most of my day-to-day commits go to private client repositories.

Open source contributions

  • roboflow/supervision (11 merged): fixed mAP false-positive counting on background images, from_tensorflow box scaling, Pascal VOC export mutating its input, and crashes on malformed Gemini/Qwen output; vectorized mask_to_xyxy and KeyPoints.as_detections with bit-identical results.
  • roboflow/trackers (5 merged): opt-in camera motion compensation for OC-SORT, a vectorized HOTA evaluator, TrackEval parity for MOT17 distractor classes, and BoT-SORT and IoU fixes. Hybrid-SORT is in review in #608.
  • roboflow/rf-detr (2 merged): the matcher now uses the configured focal_alpha, and predict() always runs in eval mode.
  • unslothai/unsloth: audit_supervision, a pre-flight report of what an SFT dataset actually supervises.
  • wasserth/TotalSegmentator: CLI capability discovery, run reports and cohort statistics.
  • openenv/OpenEnv (2 merged): env-client proxy race and provider health-poll fixes.
  • cBioPortal/cbioportal-mcp (2 merged): study-lookup and install fixes.

All merged pull requests

Projects

Project What it does
segauge DICOM-native evaluation for medical image segmentation: surface-distance metrics, per-lesion detection and a confidence interval on every number. pip install segauge
sitecheck Checks whether a medical-ML result measures biology or the lab that produced the data.
fiftyone-shortcut-risk The same shortcut check as a FiftyOne plugin: which metadata fields (site, scanner, batch) your embeddings can read and your labels correlate with.
todo-for-agents A to-do list you share with Claude Code and Codex: GitHub issues, a small CLI, and a macOS window that shows which agent works on what.
gitquake Your GitHub history as a seismograph, with AI model releases as fault lines. One stdlib-only Python file.
vitalocal-garmin A private AI health dashboard for Garmin data that runs entirely on your own laptop.

Pinned Loading

  1. gitquake gitquake Public

    Your GitHub history as a seismograph, with AI model releases as fault lines. One stdlib-only Python script.

    Python 1

  2. segauge segauge Public

    Geometrically-correct, DICOM-native evaluation metrics for medical image segmentation in Python: Dice, IoU, HD95, ASSD, Normalized Surface Dice, per-lesion detection F1, with bootstrap confidence i…

    Python 1 1

  3. sitecheck sitecheck Public

    Audit whether a medical-ML result measures biology or the lab that produced the data.

    Python

  4. todo-for-agents todo-for-agents Public

    A to-do list you share with your coding agents: GitHub issues, a CLI for Claude Code and Codex, and a macOS window that shows which agent works on what.

    Swift

  5. roboflow/supervision roboflow/supervision Public

    We write your reusable computer vision tools. 💜

    Python 51.4k 4.9k

  6. roboflow/trackers roboflow/trackers Public

    Trackers gives you clean, modular re-implementations of leading multi-object tracking algorithms released under the permissive Apache 2.0 license. You combine them with any detection model you alre…

    Python 4k 425