Machine learning research engineer. Computer Science, Morgan State University, Class of 2027. Baltimore, Maryland.
Email: johnsonkc201@gmail.com
LinkedIn: linkedin.com/in/johnsonkc
Currently seeking: May 2027 Job in machine learning research and ML engineering or Software Engineering.
I am training to be a machine learning research engineer: the person who takes a method out of a paper, implements it correctly, proves it matches the reference, and turns it into code the rest of the lab can build on.
That one goal shapes everything below. My research is in optimal transport and geometric machine learning. My open-source work goes upstream into the scientific Python stack those methods run on. My projects are where I practice getting a model out of a notebook and into something people actually run.
I graduate in Spring 2027 with a 4.0 GPA.
Machine Learning Researcher, Morgan State University, NIH NIDA R21
Advised by Dr. Pilhwa Lee. Extending optimal transport gradient flows onto symmetric positive definite matrices, applied to brain connectivity. I implemented a published sliced Wasserstein flow on the SPD manifold, validated it against our own log-domain formulation, and built the test suite that keeps both implementations honest.
Machine Learning Researcher, UC San Diego HDSI, STARS 2026
Advised by Dr. Duncan Watson-Parris, in the climate modeling group. Online bias correction for JAX-based general circulation model physics.
NLP Research Intern, Luminoso Technologies
Enterprise natural language processing and text analytics tooling.
Break Through Tech AI Fellow, Cornell Tech, Machine Learning Foundations track, 2026/27 cohort.
Most of these are libraries I use in my research, so I fix them upstream when they break. Merged:
| Project | Contribution |
|---|---|
| equinox | Treat JAX and NumPy arrays as interchangeable when deserialising |
| scanpy | Use the non-zero median as target_sum on the sparse normalization path |
| anndata | Suggest the correct store class when reading a packed zarr store by path |
| zarr | Allow require_array to accept a ZDType |
| xarray | Fix stacking of dimensions with falsy names |
| MLSysBook | Fix the broken-link count: followed redirects no longer count as broken, and DNS, TLS and timeout failures now do |
| jax-gcm | Make MoistAirColumnState broadcasting-native, make LottMillerSso work on any horizontal layout, and add optional specific-humidity relaxation for nudging |
| sqlite-utils | Honor --no-headers for formatted and table output |
| ai-job-search | Store standalone count columns as counts rather than indexes |
| OpenMontage | Key the total-failure guard on failures rather than on skips |
| yc-oss | Map unmapped company repositories, and search placeholder entries when discovering them |
Under review:
| Project | Contribution |
|---|---|
| transformers | Fix dtype cast in TimesFM 2.5 ResidualBlock for quantized weights |
| PyTorch Geometric | Fix the dense_to_sparse backward pass for three-dimensional input |
| PyTorch Geometric | Fix negative_sampling when no negative samples are requested |
| lm-evaluation-harness | Warn when metric_list names a metric the output type never emits |
| open-instruct | Document which chat template the released RL Zero checkpoints used |
Echo Flow
Local-first voice dictation for Windows. Whisper transcribes on device and a local language model cleans the text, so audio never leaves the machine. It learns your vocabulary and casing from your own corrections, supports 16 languages, and ships with more than 1,800 tests. Python, Whisper, Ollama.
pixelpets
A pixel cat or dog that lives on your desktop, watches your cursor, and reacts when you type. 14 cat coats and a Black Lab, all original art and procedural audio. Try it in the browser, no install required. JavaScript, Electron.
Yahtri
Hyperlocal resale marketplace. Listings are generated from a photo by a vision model, with map-first browsing and price-drop alerts. React, FastAPI, Supabase.
BearBoard
Student collaboration platform for Morgan State, with an LLM-backed quiz generator. I led the work that took it multi-worker on a shared Redis backplane and closed a set of critical authentication and rate-limiting issues found in review. React, FastAPI, Postgres.
Diabetes prediction from sparse EHR data
Logistic regression for early Type 2 diabetes detection in low-resource settings. Dataset, model, and write-up.
Machine learning: PyTorch, JAX, Equinox, scikit-learn, Hugging Face Transformers, NumPy, pandas, Scanpy, AnnData, Whisper
Languages: Python, R, SQL, TypeScript, JavaScript, Bash
Systems: FastAPI, React, Postgres, Redis, Docker, Git, Linux
APEx Honors at NSBE. Tau Sigma and NSLS honor societies. SABCYL Cohort 4 through the Center for Urban Families. Peer mentor for transfer students at Morgan State.
The fastest way to reach me is email: johnsonkc201@gmail.com
I am also on LinkedIn.
