United Kingdom
Give this day a chance to be your best! Most passionate about distributed systems and low-level systems programming and design, in setups where speed, resource efficiency, scalability and security are of core interest. I have an extensive background in algorithms & data structures, having participated in many national and international competitive programming competitions at high-school and university level.
• Designed and implemented an algorithm that tracks and estimates the orders' queue positions in Market By Price (MBP) data feeds using C++.
· Designed and implemented a stress-testing framework for multi-model, multi-GPU inference workloads using Python, tracking and aggregating latency and memory usage metrics. · Researched and integrated highly efficient compression algorithms for floating point time series data.
• Implemented a Temporal Convolutional Network (TCN) model that uses a single-camera view and employs a keypoint-based human skeleton. Learns to predict the keypoints’ 3D positions and rotation angles based on information from previous frames. · Improved a state-of-the-art hand pose estimation model by modifying its HRNet backbone architecture and applying contrastive training techniques to enhance invariance to rotation and scale. Improvement of average hand pose error: 6.3 mm → 5.9 mm (FreiHand dataset) • Implemented a light-weight version of the model to run on an edge device with 5x faster inference time. • Created a PyTorch dataset that uniformly cycles through non-overlapping frame sequences from multiple training videos.
• Created a web interface using React for an internal service that on-boards new Prime Video partners to Login with Amazon. • Designed the service's authorization infrastructure using AWS Cognito and internal Federated Identity Provider services. It makes use of the OpenID Connect technology, which allows obtaining information about the end-user in a REST-like manner.