Skip to content
View niharj2's full-sized avatar
5 followers · 8 following

Achievements

Achievement: YOLO

Achievements

Achievement: YOLO

Block or report niharj2

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
niharj2/README.md

Hi, I'm Nihar 👋

CS @ UIUC

Interested in the layer between ML models and the hardware they actually run on — GPU kernels, inference performance, systems work. Still early in it, and mostly learning by building things from scratch.

What I'm working on

  • CUDA SGEMM kernels — six progressively optimized matmul kernels written from scratch, 55× over naive and ~53% of cuBLAS on an H200, wrapped as a PyTorch C++/CUDA extension
  • Research at UIUC — mechanistic simulation and digital twins for cardiovascular disease (ongoing, private)
  • Previously: a retrieval system over ~90,000 research papers, and a no-code ML workflow platform running distributed training on cloud GPUs

Published

Blood Cell Detection using Deep Learning on Mobile Platforms — CSCI'23, IEEE. Trained YOLOv8 and EfficientDet to detect blood cells, then deployed them on-device to iOS and Android.

Tools I actually use

Python · C++ · CUDA · PyTorch · TensorFlow · OpenCV · HuggingFace · Docker · Linux · SLURM · GCP

Reach me

LinkedIn · niharj2@illinois.edu

Pinned Loading

  1. cuda-sgemm-pytorch-extension cuda-sgemm-pytorch-extension Public

    Six progressively optimized CUDA SGEMM kernels — 55× over naive, 53% of cuBLAS on an H200 — exposed as a PyTorch C++/CUDA extension

    Cuda

  2. Scientific-Literature-Retriever Scientific-Literature-Retriever Public

    Multi-stage retrieval over ~90k arXiv papers — BM25L, embedding search, and cross-encoder reranking

    Python

  3. Blood-Cell-Detection-on-Mobile-Platforms-using-Deep-Learning Blood-Cell-Detection-on-Mobile-Platforms-using-Deep-Learning Public

    YOLOv8 + EfficientDet blood cell detection deployed on iOS and Android — published at CSCI'23 (IEEE)

    Swift

  4. First-Tech-Challenge First-Tech-Challenge Public

    FTC robot autonomous routines — finite state machines and odometry in Java

    Java