Farzad Sabahi, Ph.D.

Leading development of multimodal, real-time end-to-end pipelines for video-based analysis. Designed high-throughput computer vision infrastructure for behavior prediction and alerting. Collaborated with researchers from multiple universities to transition AI research into deployable systems. Introduced pruning and optimization techniques that reduced model inference time by 40%. Implemented robust pipeline ensuring safe model deployment in high-stakes use cases.

Provided consulting for stock price forecasting using LSTM and Transformer models, improving prediction accuracy by 12%. Engineered features from historical data, news sentiment, and macroeconomic indicators. Delivered production-ready Python pipelines with automated data ingestion and retraining. Designed market trend detection and trading signal systems using ML to classify bullish/bearish patterns and anomaly detection for unusual price-volume movements. Created real-time data pipelines and explained model outputs, improving clients' trading decision accuracy by ~15%.

Published 15+ peer-reviewed papers on image retrieval, attention models, and generative networks. Led research on morphological networks, multi-view learning, and spatial feature aggregation for visual understanding. Developed custom deep learning architectures for large-scale visual datasets. Executed full ML pipelines including dataset preparation, model design, training, evaluation, and deployment independently. Applied MLOps principles for reproducibility and scalability. Adapted image retrieval methods to time-series market data and proposed SmartGriffin pruning strategy for inference efficiency on trading-inspired tasks.