Tom Almog

WatStreet

Financial Markets, Machine Learning

• Developing a market-regime classifier using volatility, autocorrelation shifts, and trend persistence features
• Implementing Hidden Markov Models and GARCH based signals to detect transitions between regimes
• Training a lightweight Transformer sequence model, improving regime-shift detection accuracy by over 35%
• Building a research pipeline for signal generation, backtesting, and evaluation using NumPy, Pandas & SciPy

Wat.AI

Deep Learning, Medical Imaging, Explainable AI

• Contributing to SEE-DR, a mobile AI system for early detection of diabetic retinopathy using CNN-based image segmentation
• Implementing Grad-CAM to generate interpretable heatmaps highlighting clinically relevant regions
• Preparing and preprocessing large-scale image datasets for model training
• Collaborating on model deployment pipelines using TensorFlow Lite and ONNX Runtime for mobile platforms

Waterloo Aerial Robotics Group

Machine Learning, Computer Vision, UAV Autonomy

• Training deep learning models for perception tasks including object detection, tracking, and scene segmentation
• Curating and preprocessing vision datasets to improve model generalization and robustness
• Deploying optimized ML models to UAV systems for real-time inference, integrating with navigation pipelines

myPip

AI Research & Product Development

• Developing multimodal LLM pipelines that convert user prompts into deployable React Native app components
• Improved accuracy of generated UI structures by 40% through iterative fine-tuning and model pipeline optimization
• Built modular API connectors for generated apps, supporting services like Spotify and allowing real-time data access