ML/AI
DBV
Overview
A transfer-learning image classifier trained to distinguish 120 dog breeds with high accuracy, built as a complete notebook from data pipeline to evaluation.
Approach
- MobileNetV2 backbone with transfer learning for efficient training and inference.
- TensorFlow pipeline covering augmentation, training, and evaluation.
- Clean, reproducible workflow suitable for edge deployment.
More work
View allML/AI2025
DermAI Pakistan
On-device dermatology decision support for low-resource clinics. A quantized deep learning classifier that runs entirely on the phone.
View case study
Systems/Low-level2024
ml_frame
A zero-dependency machine learning framework in roughly 2,800 lines of C11: arena allocator, reverse-mode autograd, multi-backend GEMM, and INT8 quantization.
View case study