Computer vision engineer, based in Lahore. I do some research on the side and build the tooling I end up using.
CV founding engineer at a clinical computer vision startup. I lead the annotation team, research new architectures, and run deployment and testing on edge hardware.
| Project | About | Stack |
|---|---|---|
| VoiceFlow | Local voice dictation on faster-whisper. Runs on your GPU, nothing leaves your machine. | Python faster-whisper Pyloid |
| Vision-Dissect | Cracks open CV models to compare layer activations and attention maps across YOLO11, SAM, and DepthPro. | PyTorch ONNX Transformers |
| Android-Ui-MCP | MCP server for Android UI automation and testing workflows. | TypeScript MCP |
| ultra-instinct-claude-code | 176 Claude Code tips distilled from 17 repos and 500k+ stars. Tagged by difficulty, nothing to install. | Research Docs |
Mapping Air Pollution Sources with Sequential Transformer Chaining
NeurIPS 2024 Climate Change AI Workshop. Second author.
Chained Vision Transformers with Remote CLIP to find factory and brick-kiln chimneys in South Asian satellite imagery. Filtered a 600K+ image dataset down to the ~1% that actually contained pollution sources. Paper.
LocaGraph: Learning Localized Graph Attention with Anisotropic Adaptation
NeurIPS 2025 submission. Lead author. Graph neural networks for spatial data, under review.