Mentoring a project on making GPTs using Kolmogorov Arnold Networks, a new paradigm for making deep learning models which were found to outperform Multi Layer Perceptrons (MLPs). This project was built to explore the capabilities of Kolmogorov Arnold Networks (KANs) as a deep learning paradigm and its applicability to building Large Language ModelsMentoring a project on Deep multiclass audio classification where mentees developed deep learning models using Convolutional Neural Networks, transfer learning using efficient net b0 and using pretrained models like google's Yamnet on a large dataset of audio.

Refactored existing code to optimize precompilation and loading times of over 25 various solver modules present in OrdinaryDiffEq.jl - a Differential Equation based library written in pure Julia language.Segmented complex packages into 15 manageable subpackages of solvers for Ordinary Differential Equation Solvers, improving code clarity and significantly boosting team productivity and project turnaround time.Optimized precompilation times from 2.46 seconds initially to 0.56 seconds at the end of the project which accounted for a total of 76% optimization.