Distributed cloud platform in C++17 — webmail, file storage, and user accounts backed by a fault-tolerant key-value store with synchronous replication, LRU eviction, tablet splitting, and heartbeat…

This project introduces a PDE-Aware Optimizer that aligns gradients from PDE-residual, boundary, and initial losses, scales steps by the variance of per-sample PDE gradients, and avoids expensive H…

DDPM-based U-Net transformer for fluid dynamics prediction, reproducing and extending DiffFluid. Validated on Navier-Stokes vorticity and Lattice Boltzmann (D2Q9) with corrected noise-prediction lo…

A GPU path tracer in CUDA with physically-based light transport: Lambertian, mirror, and refractive BSDFs. Features stream compaction via Thrust, stochastic anti-aliasing, depth of field, OBJ mesh …

100x faster text-guided image editing on consumer GPUs (~6s vs ~10min). Combines SSD-1B, LCM-LoRA 4-step inference, and ControlNet-Canny. Benchmarked on PIE-Bench with +23.5% CLIP score over DDIM b…

A CUDA implementation of openGJK, with GJK and EPA algorithms for massively parallel collision detection between convex polytopes. Up to 37x speedup over CPU.

Fast and reliable implementation of the Gilbert-Johnson-Keerthi (GJK) algorithm for C, C++, C#, Go, Matlab, Octave, Python and Zig