Borda - Overview

Hi there 👋 I'm Jirka Borovec

Open-source ML/CV systems maintainer. I turn research-grade computer vision and PyTorch workflows into reliable Python libraries, reproducible benchmarks, clean APIs, and contributor-friendly maintainer systems. Background: Ph.D. in Medical Imaging, former Director of OSS at Lightning AI, and long-term practitioner across ML frameworks, CV tooling, and reproducible research.

🏆 Impact Ledger

Area	Artifacts	My role	Contribution signature
Computer vision systems	supervision, RF-DETR, trackers	Maintainer in the Roboflow CV ecosystem	Model-agnostic detection, segmentation, tracking, adapters, examples, reliability
PyTorch ecosystem (emeritus)	PyTorch Lightning, TorchMetrics	Former Director of OSS / emeritus maintainer	Training infrastructure, distributed metrics, OSS governance, releases
PyTorch ecosystem (past core)	Lightning Utilities, Bolts, Flash, Thunder, Tutorials, Ecosystem CI, LitGPT	Core maintainer / contributor	Shared infra, reference implementations, tutorials, CI tooling, LLM tooling
Maintainer tooling	.github, AI-Rig, pyDeprecate, affordable-GPU-CI, cachier	Creator / maintainer	Defaults, agents, API migration, CI reliability, release discipline
Medical imaging & benchmarks	ANHIR, BIRL, pyImSegm	Researcher / organizer / library author	Histology registration, segmentation, benchmarking, reproducibility
Community & applied ML	Kaggle, Medium, tutorials, reviews	Practitioner / explainer / reviewer	Reproducible notebooks, ML practice, public technical writing

I use "creator", "maintainer", "contributor", "emeritus maintainer", and "organizer" precisely. Large ecosystem projects are team efforts. My signature is turning research-grade ideas into durable, tested, user-facing Python systems — reliable interfaces, reproducible benchmarks, safe API evolution, and contributor-friendly maintainer infra.

🛠️ Developer Track

Create & maintain several open-source Python packages used by thousands of developers.
Contributed code, CI/CD pipelines, issue reports & reviews across the ML ecosystem.
Strong focus on testing, automation, and developer experience — from pre-commit hooks to GitHub Actions.
Kaggle contributor across notebooks, competitions, and datasets — applied ML on real-world problems with published, reproducible work.

🧑‍🏫 Manager Track

Built and led a team at Kendaxa (startup, now defunct) — delivered a video-analysis platform from prototype to production.
Former Director of Open Source at Lightning AI — led the OSS team for 3+ years, driving feature roadmaps, release cycles, and cross-team coordination across PyTorch Lightning, TorchMetrics, and the broader Lightning ecosystem. Mentored contributors, scaled community engagement, and ensured quality across 10+ active repositories.
LinkedIn Learning certified in Leadership Foundations, Leadership: Practical Skills, and Leading Your Team Through Change.

🎓 Academic Track

Ph.D. in Medical Imaging — Czech Technical University in Prague.
15+ journal articles & 20+ conference papers (ISBI, ICIP, ACCV, MICCAI workshops).
Reviewer for IEEE TMI, TCIA and major international conferences.
Co-organized the ANHIR challenge on histological image registration.

🚀 Open Source & Projects

Long-term open-source contributor and maintainer. My work spans ML frameworks, developer tooling, and computer vision — always aiming to make research more reproducible and engineering more enjoyable.

Active projects I maintain:

👁️ supervision The go-to Python toolkit for plugging any detection or segmentation model into real-world CV pipelines. Unlike framework-specific tools, it works with YOLO, Transformers, or any custom model out of the box — providing a unified API for tracking, filtering, annotating, and chaining operations that would otherwise require glue code.
🎯 RF-DETR A new take on real-time object detection that brings transformer accuracy to YOLO-level speeds. Stands out by matching or beating state-of-the-art on COCO while being straightforward to fine-tune on custom datasets — no complex anchor tuning or NMS hacks needed.
🏃 trackers Plug-and-play multi-object tracking for any detection model — YOLO, DETR, or anything that outputs bounding boxes. Ships clean-room implementations of SORT, ByteTrack, OC-SORT, and BoT-SORT rebuilt from the original papers, not wrapped forks. Integrates natively with supervision.Detections for zero-glue wiring, with pre-tuned parameters benchmarked across four standard MOT datasets.
⚙️ .github Standards layer for ML/CV Python projects: default community files, templates, security policy, contributor guidance, and reusable CI/CD workflows. Shared foundation for consistent repo hygiene across projects.
🤖 AI-Rig Execution layer for AI-assisted maintainer workflows: specialist agents for code review, release audit, and human-in-the-loop quality checks. Built to keep agents calibrated and maintainers in control.
♻️ pyDeprecate Born from the pain of managing API changes in large libraries like PyTorch Lightning. A zero-dependency tool that lets library authors deprecate, rename, and redirect functions or classes with automatic call forwarding — so users get clear migration warnings instead of silent breakage.

Emeritus maintainer — projects I co-created and still partially supervise:

Past core maintainer projects:

🛠️ Lightning Utilities The shared foundation that keeps all Lightning projects consistent and maintainable. Extracts common patterns — packaging helpers, testing utilities, CLI tooling, and CI/CD workflows — into one place so that fixes and improvements propagate across the entire ecosystem automatically.
🔩 Lightning Bolts A community-driven collection of reference implementations — VAEs, GANs, SimCLR, and more — built on PyTorch Lightning. Designed to give researchers battle-tested baselines they can reproduce in one command and extend for their own experiments.
⚡ Lightning Flash Made transfer learning as simple as a few lines of code across 15+ tasks — image classification, object detection, text classification, tabular data, and more. Built on PyTorch Lightning, it let practitioners go from idea to baseline in minutes instead of hours.
🌩️ Lightning Thunder A source-to-source compiler for PyTorch that delivers up to 40% faster training and inference through kernel fusion, operator optimization, and GPU memory management. Unlike opaque compilers, Thunder provides a transparent, Pythonic IR that developers can inspect and customize — with composable plugins for distributed training, quantization, and CUDA Graphs.
📚 Lightning Tutorials The official tutorial collection powering the PyTorch Lightning documentation. Uses a script-based format instead of heavy notebooks — automatically converting to executable notebooks with full reproducibility tracking, CI-tested across CPU, GPU, and TPU to ensure every example actually runs.
🔄 Ecosystem CI The safety net for the entire Lightning ecosystem — automatically runs downstream test suites against every nightly build and release candidate. Catches breaking changes before they ship, ensuring that hundreds of dependent projects don't break on upgrade day.
🧠 LitGPT An opinionated, hackable codebase for working with 20+ LLMs — GPT, Llama, Mistral, and more. Unlike heavyweight frameworks, LitGPT uses plain PyTorch with no abstraction layers, making it easy to modify any part of the training pipeline while still getting optimized performance out of the box.

Past research projects:

🖼️ pyImSegm A complete image segmentation pipeline developed during Ph.D. research, combining superpixels, graph cuts, and region growing for medical imaging. Used in multiple published studies on histological tissue analysis and designed to be reproducible from raw data to final results.
📊 BIRL The benchmarking engine behind the ANHIR grand challenge at ISBI, which brought together teams worldwide to compare image registration methods on histological data. Automates the full pipeline from running registration to evaluating alignment accuracy using expert-annotated landmarks.

Notable contributions to other projects: ultralytics/YOLOv5, DIPY and more...

📊 GitHub Stats

🏅 Kaggle Stats

💖 Support & Consulting

If my open-source work is useful, consider sponsoring me 💚

Signature support areas — see SUPPORT.md for details:

OSS Health Audit — repository structure, CI, release process, docs, issue/PR flow, contributor experience, and maintenance risk.
API Lifecycle & Migration Sprint — versioning strategy, deprecation policy, migration warnings, compatibility layers, tests, and release notes for Python libraries.
Computer Vision Pipeline Reliability Sprint — detection, segmentation, tracking, adapter, fixture, evaluation, and reproducibility gaps in CV systems.
AI-Assisted Maintainer Workflow Setup — .github defaults, AI-Rig-style review workflows, release checks, PR Evidence Cards, and human-in-the-loop review for Python/ML OSS teams.

I do not sell generic AI adoption. I help teams make ML repositories maintainable, testable, and release-safe.