I build computer vision systems: multimodal and vision–language models that bring domain knowledge into detection, segmentation and scene understanding.
Graduating in 2027 and open to applied and research scientist roles.
I'm a PhD candidate in Computer Science at the University of Virginia, advised by Madhav Marathe at the Biocomplexity Institute. My PhD research is in computer vision for Earth observation: knowledge-guided and multimodal learning, vision–language models, and foundation-model adaptation for remote sensing, applied to agricultural and environmental monitoring. I design models that fold domain priors into segmentation and detection, and I build the large-scale, ML-ready geospatial datasets that make those models trainable and testable.
I'm currently an Applied Scientist Intern on Amazon's GRAISE team in Seattle, working on multimodal computer vision models for grocery. Before joining Amazon I spent a summer as a Graduate Research Intern in Oak Ridge National Laboratory's GeoAI group, and before the PhD I worked as a software engineer at Enosis Solutions. Work from my PhD and internships has appeared at NeurIPS, KDD, IJCAI, WACV and CVPR workshops.
Segmentation and detection models that take domain structure as input instead of relearning it: crop and land-use priors through cross-attention, graph constraints on canal networks, and geometric priors on infrastructure.
Scalable pipelines that turn raw satellite and UAV archives into annotated, multimodal benchmarks with metadata, held-out sites and task-specific metrics.
WACV 2027 accepted
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Datasets Track First public benchmark for fallen-log detection in centimeter-resolution UAV orthomosaics: 10,399 verified instances with oriented boxes and masks, paired LiDAR and photogrammetric height models, and a centerline-based metric for elongated objects.
OASIS @ SIGSPATIAL 2026
1st ACM SIGSPATIAL Student Challenge on Open Agents with Spatial Intelligence for Social Good (OASIS 2026) Challenge finalist. A hybrid agentic system that pairs LLM agents for location, structure and reasoning with deterministic geospatial tools for flood assessment, facility discovery and evacuation routing.
EarthVision @ CVPR 2026
EarthVision Workshop, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
WACV 2026
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
NeurIPS 2025
Advances in Neural Information Processing Systems (NeurIPS)
IJCAI 2025
International Joint Conference on Artificial Intelligence (IJCAI)
IJCAI 2025
International Joint Conference on Artificial Intelligence (IJCAI)
