I led the development of a high-throughput, multi-model intelligent visual analytics system. Besides the core architecture and low-latency frame-processing implementation, I also implemented and productionised state-of-the-art machine-learning models including detection of people and faces, body and face recognition, tracking incorporating a Kalman Filter, demographics and pose extraction. Additional unique capabilities of the system included a novel whole-system calibration process, and the 3D math to fuse all camera inputs into a cohesive tracked space.Additionally, I was the primary technical salesperson for international business and main contributor of technical prose to proof-of-concept documents and investor/customer pitch decks.I led the hiring for Machine Learning Engineers, attracting and training a strong team with deep technical ability in performance tuning, research productionisation and model building. I also led our deep learning hardware specification and acquisition, improving productivity and decreasing cost across the group.Visual analytics models and technologies:- gstreamer, deepstream and Pytorch, CUDA streams- body and face detection (SSD300, FasterRCNN, RetinaFace, EXTD_64, MTCNN)- body and face identity (OmniScale ReID, ArcFace, FaceNet etc)- single-camera tracking based on DeepSORT, extended for multi-camera contiguous spaces- demographics (Deep Label Distribution Learning jointly trained for age, gender, ethnicity)- pose detection (PifPaf, and Keypoint FasterRCN w/ FPN)Recommendation system:- rebuilt production-ready data pipeline- designed and implemented sequence-based user-personalisation deep learning model in Pytorch- built highly efficient on-GPU business rules engine

- Built customer-next-action LSTM-based sequence model in Pytorch, deployed to AWS Elastic Container Service.- Implemented collaborative filtering and content models, and a stacked ensemble recommendation model.

Gluent successfully attracted many Fortune-500 customers (including several in the Fortune-50), with consistently successful deployments and high stakeholder satisfaction. The Gluent Data Platform is a big-data middleware that connects modern Hadoop-based data sources with established enterprise databases, unlocking value and increasing efficiency.- Built the first version of the full product, including high-efficiency C++ data pipeline connector, and Python-based CLI data orchestration tools.- Hired, trained and managed the teams (8 hires) that would a) develop the software to enterprise production grade, and b) implement the solution at many Fortune-500 clients.- Advised CEO on product development, timelines, business opportunities and strategy.

Developed "Transportist" - the first public transport application for Istanbul, Turkey. This was also the first application worldwide to perform trip planning on the device itself without requiring a network connection.Coverage later expanded to around 40 cities throughout Europe, the United States and Asia.

Implemented pioneering voice communications app on iOS. Users could communicate with high quality audio messages, and these could be sent asynchronously with a highly streamlined user interface.

Implemented full-stack web services and backend agents for enterprise Capacity Planning and Server Estate Management product. Coffeescript, Python/Django for the web stack and Scala for the backend.

Front-end trading system development in C++, C# and Java. C++ hiring.