2018 — 2023
Ph.D. in Computer Science
Ben-Gurion University of the Negev
Dissertation: Using Weakly Supervised Learning to Solve Visual Computing Problems.
Personal record · v2026
Ahmad Droby
Ph.D. in Computer Science · Computer Vision & Machine Learning
I lead the algorithms team at Neolithics.ai, building production computer-vision and hyperspectral models for fresh-produce quality control. Before that, I co-founded Mirage Dynamics as CTO and completed my doctorate at Ben-Gurion University on weakly-supervised learning for visual computing.
01
Education
Degrees & awardsComputer science at Ben-Gurion University of the Negev, Beersheba — from BSc through to a doctorate on weakly-supervised learning for visual computing problems.
Timeline
2016 — 2018
M.Sc. in Computer Science
Ben-Gurion University of the Negev
Thesis: Surface Detection and Deformation Detection in a Video Stream.
2013 — 2016
B.Sc. in Computer Science
Ben-Gurion University of the Negev
2016
Android & Web Application Development
The College of Management, Beersheba
02
Experience
Roles & engineeringProduction ML for fresh-produce inspection, AR-style in-video advertising, and academic teaching. Stack is PyTorch / Lightning / Optuna for training, ONNX / TensorRT for inference, and a lot of careful data engineering.
Roles
Algorithm Team Lead
Neolithics.ai
2025 — Present
- Designed and led the Automatic Model Build Flow — data collection, dataset lifecycle, model building, deployment.
- Developed advanced computer-vision and hyperspectral models for blueberries, avocados, grapes and garlic.
- Built multi-label and ordinal classification heads (CNN, ViT, CNN-ViT hybrids) with severity-aware losses.
- Established reproducibility and deployment standards: TorchScript / ONNX / TensorRT.
- Engineered large-scale training pipelines with PyTorch Lightning and Optuna HPO.
Senior Computer Vision Researcher
Neolithics.ai
2024
- Led R&D turning hyperspectral and RGB imagery into real-time, production-grade quality predictions.
- Designed a multimodal fusion network blending RGB and 224-band hyperspectral data.
- Built end-to-end ML pipelines: ingestion, feature extraction, automated validation.
- Delivered faster Jetson-ready inference via hybrid CNN / transformer + signal-processing techniques.
CTO & Co-Founder
Mirage Dynamics
2022 — 2024
- Co-founded Mirage Dynamics and led integration of AR-style in-video advertising using AI.
- Built a system for analysing and replacing existing ads in videos — detection, segmentation, tracking.
- Dynamic ad insertion for live HLS streams, optimised for high performance.
- Contextual video analysis using image/video captioning, NLP and zero-shot classification.
- Hybrid AI + Adobe After Effects pipeline for premium-quality ad insertions.
03
Publications
Selected papers & datasetsFive selected items below. Full list and citation graph on Google Scholar and DBLP.
Selected
2022
Digital Hebrew Paleography: Script Types and Modes
A comprehensive study applying deep learning to classify medieval Hebrew manuscripts into 14 classes based on script style and mode.
2022
Textline Alignment on the Image Domain
A novel approach to text-line alignment in document images, improving downstream recognition accuracy.
2022
HST-GAN — Historical Style Transfer for Text Images
Generative model for synthesising authentic-looking historical text images via style transfer.
2021
Unsupervised Text-Line Segmentation by Differentiating Coarse Patterns
Unsupervised approach to text-line segmentation in handwritten documents.
2021
VML-HD — Historical Arabic Documents Dataset
Dataset for advancing Arabic historical document recognition research.
04
Contact
Get in touchOpen to collaboration on computer-vision research, hyperspectral imaging projects, and production ML for industrial inspection.