AI/ML Engineer · Researcher · Toledo, Ohio
Making machine learning work on signals that don’t behave
I build ML systems for physical processes — furnaces, hip scans, roads, microscopes — where the measurement is as hard as the model. I’m an ML engineer at the Northwest Ohio Innovation Consortium, partway through a master’s in computer science engineering at the University of Toledo.
- 3 Published papers Springer Nature · IEEE · UPenn-validated
- 1 Granted patent Flower Plucking Robotic Arm, IN 573184
- $50,000 NSF grant I-Corps, osteoporosis risk tool · patent pending
- 6 Featured projects industrial · medical · transport
About
I work on the machine learning that has to survive contact with hardware.
I build machine learning systems for physical processes — furnaces, hip scans, roads, microscopes — where the measurement is as hard as the model. The through-line is hybrid fuzzy–deep learning: a fuzzy layer for the ambiguity, a deep model for the structure underneath it.
Glass furnace · Northwest Ohio Innovation Consortium · 4.37°F
The sensor fails. The reading continues.
Thermocouples in molten glass drift, degrade and fail — and they measure the variable that sets fiber diameter, breakage and energy cost.
Six gradient-boosted models read underglass temperature from the crown instead, at 4.37°F RMSE mid-melt at 100 TPD.
Osteoporosis · MS thesis research, University of Toledo · 88%
Risk, from a scan you already had.
I isolate bone from a routine CT, extract 18 density features and combine them with age, BMI and lab markers — high, medium or low risk in minutes, at approximately 88% accuracy.
It reads imaging patients have already had, instead of the separate DEXA appointment screening normally depends on.
Approximately 88% is the university’s figure, reported in UToledo News, 8 September 2026.
Brain tumor · University of Toledo, Dept. of EECS · 95.8%
Where the boundary is not a boundary.
Tumour margins are soft and blurred, which is exactly where hard clustering fails. Fuzzy C-Means segments the ambiguous boundary and a CNN classifies what it finds.
Cell death · University of Toledo
Reading cell death without a stain.
Studying how cancer cells die normally requires fluorescent labelling, which is invasive and can alter the biology being observed. This asks whether death patterns can be recognised from raw, unlabelled microscopy video alone.
Traffic forecasting · Ohio Department of Transportation · 8.92%
Seven years of counts is not a time series.
With only seven years of counts, time-series forecasting is the wrong frame. Normalising the origin-destination matrix by population turns the task into learning a spatial assignment operator that can be re-expanded against any year.
Speed estimation · Independent work, then Ohio Department of Transportation
Speed, from a camera that only sees boxes.
Built first independently — roadside video capture, detection, tracking and speed derived from tracked movement, with real-time inference on a Raspberry Pi at the roadside. Then integrated into ODOT’s existing statewide detection network.
Tech stack
The tools I reach for
Fuzzy–deep learning at the centre; the rest is what it takes to get a model in front of someone who can act on it.
learning
- Python
- TensorFlow
- Keras
- Scikit-Learn
- OpenCV
- FastAPI
- React
- pandas
- NumPy
- Git
- Vite
- Matplotlib
- Raspberry Pi
- MATLAB
- JavaScript
Contact
Get in touch
I’m looking for full-time ML engineering and research roles from Summer 2027, and for internships before then. The fastest way to reach me is email.