Explored and created a web-hosted demo for a zero-shot deepfake audio generation model to generate audio from text using embeddings of short reference audio clips
Explored distributing model components across different devices to enable running large models on edge with multiple smaller devices
Analysed and predicted the behaviour of visitors in NYC's Bryant Park, building a digital twin spatial model for 4D simulations
Built and deployed distributed edge compute on edge with fractional device usage for higher device efficiency using load balancing and micro kubernetes engine
Predicted demand using sales figures, mobility data and several demographic and footfall variables
Designed end-to-end biometric system with user registration and verification as a replacement for fingerprint technology
Explored Active Learning with different querying strategies on CIFAR10 dataset and managed to achieve high accuracies with very limited training data
Implemented and optimised multiple live models (eg. action recognition on something something dataset with over 200 classes) on edge devices (eg. Jetson TX2, Raspberry Pi)