Ahmad Droby

• Designed and lead the Automatic Model Build Flow: data collection, dataset lifecycle, model building and deployment as one reproducible pipeline.
• Advanced CV and hyperspectral models for blueberries, avocados, grapes and garlic; multi-label and ordinal heads (CNN, ViT, hybrids) with severity-aware losses.
• Reproducibility and deployment standards: TorchScript / ONNX / TensorRT, Lightning + Optuna training pipelines, versioned datasets on DuckDB + GCP + zarr.

• Multimodal fusion network blending RGB with 224-band hyperspectral data for real-time quality prediction.
• End-to-end ML pipelines (ingestion, feature extraction, automated validation) and Jetson-ready inference via hybrid CNN / transformer models.

• Built the system that analyses and replaces existing ads in video (detection, segmentation, tracking); dynamic insertion for live HLS streams; co-inventor on the patent application for video advertising signage replacement (WO 2021/090328).
• Contextual video analysis with captioning, NLP and zero-shot classification; hybrid AI + After Effects pipeline for premium insertions.

• Taught and assisted computer-science courses; Teaching Excellence Award, 2022.

Dissertation: Using Weakly Supervised Learning to Solve Visual Computing Problems. Advisor: Prof. Jihad El-Sana, Visual Media Lab.

Thesis: Surface Detection and Deformation Detection in a Video Stream.