ProjectsHighlightsParallel Tridiagonal Solvers for Neural Networks (Jun-Aug 2025)First reproducibility study of DeepPCR with cross-platform extensions and custom Triton kernels to validate performance claims and establish computational boundariesAMD AI Sprint Hackathon (Jul 2025)Optimising Mixtral-8x7B inference performance on AMD MI300x with vLLM v1ViT for ERA5 Weather Classification (Jan-Mar 2025)Vision Transformer (ViT) implementation with PyTorch Distributed for weather classification across multiple NVIDIA V100s, achieving up to 28x speedupDistributed Brain Simulation (Feb-Mar 2025)C++ brain simulation using event-based coordination with MPI, featuring a custom type-safe coordination framework with handler-based dispatching and batch processingmicro:bit Raycaster (Dec 2023)Textured raycaster on the NRF52833 SoC supporting collision detection and spritesOther Cool ProjectsSAC Agent Implementation (Mar 2025)Soft Actor-Critic (SAC) agent for continuous control in highway driving scenarios using the HighwayEnv environmentDeep Learning for Predicting CT Slice Locations (Oct-Nov 2024)From-scratch implementation of Gaussian Processes and Deep Neural Networks for predicting relative CT slice positions on the axial axisInland Empire (Aug 2024 - Nov 2024)Movie recommendation system based on scraped Letterboxd user dataFDASurfaces.jl (Aug 2023 - Jul 2024)Core contributor of a CUDA-powered Julia package, used for functional data analysis on smooth manifoldsSDN Controller (Feb-Mar 2024)Software-defined networking (SDN) controller using Ryu and OpenFlow with firewalling and pre-installed flow modsPath Planning for Autonomous Robots (Nov-Dec 2023)Genetic algorithm-based path planning optimisation with simulated annealingMojo Compiler (Feb-Jun 2023)Fully functioning compiler for an object-oriented language; written in JavaNine-Point Stencil Solver (Mar-May 2023)Parallelised 2D advection solver using MPI, OpenMP and CUDA; written in CConvolutional Neural Networks for Domain Adaptation (Apr 2023)Fine-tuned ResNet-18 for domain adaptation on the DomainNet dataset in Pytorch