ML engineer · medical imaging and clinical NLP · Lagos, Nigeria
I am a machine learning engineer and researcher. I build medical imaging and clinical language pipelines, and I test them hard enough to know which results to trust.
Flagship work · Principal Investigator · 2026
Most medical AI has never been validated on African patients. I lead a research programme at Federal Medical Centre Ebute-Metta, Lagos, that builds the dataset and benchmark needed to do it.
A pipeline that turns free-text radiology reports into structured labels. It combines a curated phrase dictionary with negation handling and a fine-tuned BioClinicalBERT model, joined in an ensemble. I also wrote the research protocol, the de-identification workflow and the clinician review process.
Reviewers questioned the evaluation, so I rebuilt it. I grouped reports by patient and by identical text to stop leakage, moved to nested cross-validation with thresholds chosen inside each fold, and then re-checked every claim in my draft. Twenty did not hold up, and I corrected or withdrew them.
An off-the-shelf labeller treated hedged wording, such as a heart size that could not be assessed, as a positive finding. I worked with a radiologist to write a labelling convention for uncertain language, so the dataset records what the report actually says.
Ethics committee protocol, de-identification before research use, and audits of that de-identification. Clinicians review the labels. Blinded validation by two independent raters is being set up.
Status: two papers in revision after peer review. The research programme sits within Batlak AI Lab, which I co-founded.
Other projects
Designed U-Net segmentation and CNN detection models to measure head circumference, abdominal circumference and femur length from ultrasound video. Validated outputs against clinicians' ground-truth measurements and revised the models from their feedback.
Contributed to a platform with a custom DICOM desktop viewer and cloud infrastructure. Wrote Python and Selenium automation to curate imaging datasets from PACS, converted data to YOLO and COCO formats, and worked on model development, REST API integration and inference deployment.
Built a pipeline to extract and anonymise radiology data from PACS and curate a labelled chest X-ray dataset of more than 10,000 images. Trained CNN models for multi-label pathology classification and integrated them into a diagnostic triage tool.
How I work
I check for leakage, unstable thresholds and claims the data cannot support before I report a number.
I have worked beside radiologists since 2022. I write conventions with them and build review into the pipeline.
I audit my own data handling and results, write down what I find, and correct it.
Contact
I am open to remote roles in medical imaging, clinical NLP and model evaluation, full-time or contract.