Jonathan Muhire Jonathan Muhire

About

I build software for systems that need to survive real execution: robotics simulation, CI reliability, ML evaluation, and data infrastructure. I care about tight feedback loops, reproducible proof, and debugging from logs instead of assumptions.

Recent work includes an upstream openpilot MetaDrive CI PR with 20 consecutive full workflow passes, a ROS2/C++ robotics data-capture prototype, and open-source document AI work through GSoC 2025. I am completing concurrent Computer Science and AI degrees at Oklahoma Christian University.

Current Focus

  • openpilot MetaDrive CI - Linux GitHub Actions simulator coverage, CI-scoped reliability fixes, log artifacts, and 20/20 consecutive workflow proof on the submitted PR.
  • Memory-augmented agents — ReAct with 4 memory modes, needle-in-haystack evals showing 0% to 100% retrieval recovery.
  • Document AI — GSoC 2025. Layout detection, OCR, and post-correction on Renaissance manuscripts.
  • Data infrastructure - Distributed MinIO + LakeFS for reproducible ML experiments.
  • Biomedical NLP — Token-level and span-level clinical concept extraction benchmarks.

Other Work

Stack

Python, C++, SQL, TypeScript. PyTorch, Transformers, OpenCV, ROS2, Docker, GitHub Actions, MinIO, LakeFS, PostgreSQL.

Contact

muhirejonathan123@gmail.com
GitHub · LinkedIn · Book a call