I will build computer vision and predictive machine learning models

A model that scores 95% in a notebook and 60% on your real data isn't a model - it's a leak in the validation split.

I build models that hold up outside the notebook, and report the numbers that matter, not the one that looks best.

I'm an ML engineer with 4+ years in production ML. Shipped: radiographic analysis at 95% accuracy that cut diagnosis time 40%, CNN classification at 92% with TensorRT, and an XGBoost churn model at 87%.

WHAT I BUILD

Image classification, object detection, segmentation, OCR

Defect and quality inspection

Churn, demand and risk prediction

Time series forecasting and tabular regression

HOW I WORK

Per-class metrics and confusion matrices, not one number

Proper held-out validation - no leakage, no train-on-test

Augmentation and transfer learning when labelled data is limited

Inference optimization for hardware you can afford

An honest answer if your data can't support the accuracy

STACK

PyTorch, TensorFlow, Scikit-Learn, XGBoost, OpenCV, YOLO, ONNX, TensorRT, FastAPI, Docker, AWS/GCP/Azure

Message me with what you're predicting and what data you have. If the dataset won't support it, I'd rather tell you now than after you've paid.