Hi, I'm Chinmay

I build high-performance systems and optimize the impossible

                    $ whoami

                    Software Engineer @ Venture Development Center, UMass Boston

                    CUDA Kernel Developer

                    Open Source Contributor

performance_optimizer.py

97%

Cache Reduction

FastMQA CUDA Kernel
Enabling 32x larger batches

4.06x

Speed Improvement

PyTorch AutoTune
12s → 3s per epoch

1000+

Downloads

Open Source Packages
Growing daily on PyPI

858

FPS @ 1M Points

3D Point Cloud Viewer
C++/OpenGL optimization

85%

Cost Reduction

AI Agent Platform
Smart LLM routing

300+

Active Users

Chrome Extension
4.8 star rating

Featured Projects

FastMQA CUDA Kernel

Multi-Query Attention implementation achieving 97% KV-cache memory reduction for LLM inference. Enables 32x larger batch sizes through hardware co-design.

⚡ 70,810 tokens/sec

📊 1.8x faster

🔧 CUDA/C++

View on GitHub →

PyTorch AutoTune

Published PyPI package that automatically optimizes any PyTorch model with zero configuration. Just one line of code for 4x speedup.

📦 1000+ downloads

🔋 36% energy saved

🐍 Python

View on PyPI →

View All Projects →

Technical Arsenal

Systems & Performance

CUDA C++ OpenGL WebGL

AI & Machine Learning

PyTorch LangChain Transformers ONNX

Web Development

React TypeScript NestJS WebSockets

Infrastructure

Docker Kubernetes AWS PostgreSQL

Journey

Software Developer

Venture Development Center | 2024 - Present

Building AI agents processing 50+ daily queries with 85% accuracy. Reduced LLM costs by 85% through intelligent routing.

MS Computer Science

UMass Boston | 2023 - 2025

Focusing on GPU optimization, systems programming, and distributed computing.

Software Engineer

Landmark Ventures | 2022 - 2023

Built React property analyzer enabling 3x faster deal evaluation. Automated reporting from 8 hours to 1 hour weekly.

Open Source Contributor

vLLM, PyTorch & Hugging Face | Ongoing

Open PR to vLLM (#52841: ERNIE 4.5 tool & reasoning parsers). Submitted PRs to PyTorch, including #158985. Published models with 131+ downloads on Hugging Face Hub.