Computer Science + Statistics at UC Irvine (GPA 3.85/4.0, Expected June 2028). Researching GPU systems optimization at MIT CSAIL, robot perception at CMU, and biomedical data infrastructure at Oregon State.
A brief introduction
Hi, I'm Ishaani, a Computer Science major with a Statistics minor at UC Irvine (GPA 3.85/4.0). I'm passionate about GPU optimization, high-performance computing, and applied ML — especially the intersection of systems efficiency and practical impact.
Currently researching CUDA kernel optimization at MIT CSAIL, building pose-estimation pipelines at CMU Human Sensing Lab, and architecting biomedical data infrastructure at Oregon State University.
I love bridging hardware-level performance with high-level AI to create solutions that scale. Recently won Best Overall at IrvineHacks 2026 for Candid (civic-tech platform with RAG + LLaMA), architected 90%+ accuracy computer vision pipelines, and shipped production AWS infrastructure serving $100K+ revenue.
Where I've worked and researched
Recent work and experiments
Matrix multiply from scratch across 4 optimization levels: naive, shared-memory tiling, register blocking, and warp primitives. Reaches 86% of cuBLAS at large sizes. Profiled with Nsight Compute.
ResNet-18 on CIFAR-10 (60K images, 10 classes) with data augmentation, cosine LR scheduling, and early stopping. 90% test accuracy. Deployed with ONNX for hardware-agnostic inference.
A hands-on toolkit for profiling and optimizing PyTorch data loading pipelines. Demonstrates how to cut data loading overhead from ~40% to ~10% of total training time using smart batching, parallel augmentation, and CUDA prefetching.
Turns plain English into Google Calendar events. "Study for 2 hours" → optimal slot with no conflicts, auto-synced. Reads real calendar context so there's never a double-booking.
ML extraction pipeline for aviation compliance docs with 94%+ compliance accuracy, Lambda cold-start cut 65% (8 s to 2.8 s). Serving 2 international clients processing 500+ documents/month.
Matchmaking platform for startups and investors with AI semantic search. Led 3-engineer team to win 1st place out of 11 teams at C2S Technologies' internal pitch competition ($10,000 funding).