I turn raw imagery into decisions. I build deep-learning and computer-vision systems for terabyte-scale sensor data — from foundation-model segmentation to multimodal and temporal modeling — scaling experiments on HPC and turning them into reproducible pipelines. I’m looking for roles where real-world data, computer vision, and ambitious model building meet.
I’m a Ph.D. candidate at the University of Florida, graduating in December 2026, building applied AI and computer-vision systems for large-scale, real-world sensor data — terabytes of aerial imagery, imperfect labels, shifting seasons, and models that have to hold up outside the lab.
I've worked hands-on with the whole sensing stack: RGB, multispectral, thermal, and hyperspectral cameras, flown on UAV platforms from DJI to Freefly and Inspired Flight — owning everything from sensor integration and flight operations to data processing, model training, and validated predictions people act on.
I use foundation models such as SAM and Depth Anything, object detection, multimodal fusion, and temporal models — wrapped in automated, reproducible pipelines and trained on UF’s HiPerGator HPC (SLURM).
I’m most energized by messy-data problems that demand both research judgment and engineering discipline. That opens the door to applied ML, computer vision, research engineering, multimodal AI, and foundation-model teams — including opportunities to grow toward larger-scale and world-model research.
+ DSSAT crop-modeling workshop — process-based simulation to complement the data-driven side.
A multimodal ML framework that combines UAV imagery, phenotypic traits, management records, and environmental time series to predict plant nutrient status. I built and evaluated models spanning classical ML, deep image fusion, and temporal sequence learning, then tested the final pipeline on a fully independent growing season. The work revealed where the strongest predictive signal actually comes from and how well the system holds up across seasons.
A regulation-compliant, modular drone payload integrating RGB, multispectral & thermal sensors for simultaneous capture — plus the pipeline that turns raw flights into canopy height, coverage & temperature traits, validated against ground truth.
Applied SAM (Segment Anything Model) and fine-tuned YOLO to segment individual fruit, seeds, and leaves from field imagery — cutting manual annotation and turning raw photos into structured, labeled datasets.
A multistage, automated Python pipeline turning raw drone flights into per-plot traits: orthomosaic rendering, spectral-index math, plot segmentation, canopy-height modeling & trait export — batched across thousands of plots on HiPerGator HPC.
A detection system that automatically locates strawberry runners across ground & aerial imagery, trained on a diverse multi-platform dataset for robust field-condition performance.
Segmentation + regression pipelines (watershed delineation, temporal growth-curve fits) linking aerial canopy traits to tuber weight and yield — high-throughput phenotyping replacing slow manual measurement.
A growing shelf of little characters.
Behind the lens.