Avisha Bhiryani

Achieved increased engagement from non-technical users by conceptualising and deploying a GenAI project pipeline to identify and address issues that cause anomalies in building energy consumption time series data.

Developed and deployed, on Azure ML, a multi-stage prediction pipeline for forecasting frequency deviations in power grids.

Achieved a 5% improvement in accuracy by expanding the image training dataset and implemented components in C++ to integrate the object detection pipeline.

Developed an automated real-time analytics module for an online advertising client, reducing manual reporting time by 40%.

Thesis: Enhanced robustness and policy generalisation of a reinforcement learning agent on a power grid environment with an action space of 70k using the Soft Actor-Critic (SAC) algorithm with custom observation perturbation techniques. Accelerated training by 60% by parallelising SAC on HPC clusters.