I'm an AI/ML engineer working across computer vision, remote sensing, and language models. I've trained segmentation and detection models on more than a dozen architectures, shipped geospatial ML pipelines used in production, and published research on transformer-based wildfire detection.
Adapted Meta's Segment Anything Model for prompt-based, multi-class segmentation of abdominal MRI organs — freezing the encoder and training a custom decoder for the CHAOS dataset.
Built an assessment pipeline for disaster recovery teams — trained YOLO on labeled UAV datasets and generated damage hotspot maps for on-the-ground planning.
Framework for fire segmentation across satellite imagery, benchmarking ViT-TransUNet, MedT, and U-Net. Presented at CVCI 2025 and published in ACM Proceedings.
Built and deployed a production ML pipeline that lets Tokha Municipality staff query GIS databases in plain language, translated to SQL via LangChain and LangGraph.