Skip to content
View amapara27's full-sized avatar

Highlights

  • Pro

Block or report amapara27

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. 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
amapara27/README.md

Aayan Mapara

currently working on ai tools, exploring inference, and researching ml systems + compilers.
check out my projects below, but also head to my website to learn more about me.

Projects

Anchor
Rust · Tauri · React · Ollama · MLX

find, fit, and benchmark local models on Apple Silicon.

  • indicates whether a model fits your Mac before you download it
  • measures decode speed and time to first token over multiple runtimes based on researched methodology

Jev Pilot
Python · Nemotron Speech · Jev

control your Mac with spoken commands. 300+ downloads

  • transcribes speech locally, then Jev picks a grounded action
  • checks each result before continuing to the next command

ScribeShield
Python · PyTorch · Whisper (LoRA) · FastAPI

healthcare phone-call transcription that flags uncertain words and corrects only with evidence.

  • 1st place at VillageHacks 2026; I led a team of four and built the fine-tuned model and serving backend
  • LoRA fine-tuned Whisper Small: 4.14% word error rate

GPU Kernels
CUDA · C++ · FP16 · Nsight Compute · Nsight Systems

CUDA kernels written from scratch, each checked against a reference and profiled.

  • FlashAttention-2: 2.66 ms at N=4096, 4 heads, 6.8× faster than my first multi-head version
  • SGEMM: 90.9% of cuBLAS performance
  • Reductions: a progression of CUDA reduction kernels

Pinned Loading

  1. anchor anchor Public

    Find, fit, and benchmark local models with ease.

    Rust 2

  2. jev-pilot jev-pilot Public

    Control your MacOS desktop smoothly. Powered by Jev.

    Python

  3. recon-ethereum-security recon-ethereum-security Public

    AI-powered Ethereum security platform for fraud detection and smart contract auditing

    JavaScript

  4. flash-attention flash-attention Public

    building flash attention from scratch

    Cuda

  5. sgemm sgemm Public

    SGEMM from scratch in CUDA

    Cuda

  6. sundevilmotorsports/NondimensionalTractionControl sundevilmotorsports/NondimensionalTractionControl Public

    Using the Nondimensional Tire Model to predict driven axle wheel slip and activate TC in advance

    C