I will build a custom computer vision model in pytorch

Sri Lanka

I speak Sinhala, English

ML Engineer Who Ships AI Models to Production with PyTorch and AWS

I build AI that works outside the notebook. HoloPaint Cloud, my latest project, runs a U-Net I trained from scratch as a serverless pipeline on AWS, compressed 4x with ONNX INT8 at 99% output parity, ...
About this Gig

Need a model that can look at images and sort, grade or flag them? I build custom computer vision models in PyTorch, trained on your images and tested properly.


Recent work:

  • A skin-condition severity classifier using ResNet-50 and EfficientNet features (91% test accuracy, 0.963 ROC-AUC)
  • A U-Net edge-detection model trained from scratch, now running in production on AWS
  • Real-time hand tracking and gesture control from a webcam (HoloDraw)


Every package includes:

  • Image cleaning, augmentation and a proper train/validation/test split
  • A trained model with accuracy, confusion matrix and ROC curves
  • A clean, commented notebook you can re-run
  • The model file and code to make predictions


Higher packages compare several backbones to pick the best, explain results in plain language, and add a Streamlit demo app plus an ONNX export so the model runs fast on a normal CPU.


Message me with a few example images and what you want the model to decide. I'll confirm it's a good fit before you order.

Expertise:

Image processing

•

Feature learning

•

Classification

Programming language:

Python

•

SQL

•

Colab

Frameworks:

Keras

•

PyTorch

•

Panda

Tools:

Jupyter Notebook

•

OpenCV

•

TensorFlow

•

Colab

My Portfolio