Jędrzej Maczan
Advancing machine learning research at the intersection of math, AI, and low-level systems, with a focus on LLM inference and ML compilers.
At a Glance
- AI Researchers
- Open Source Community
- ML Engineers
AI Tools by Jędrzej Maczan
(1)
tiny-vllm
LLM Inference Engine From Scratch
Discussions
No discussions yet
Be the first to start a discussion about Jędrzej Maczan
Latest News
Hosting Online Softmax session at Cohere Labs Open Science Community
Published paper on numerically stable online softmax on GPU
Released 'The cuBLAS transposition trick' research
Solved 0/1 Knapsack problem with sliding window in Paged Out! and PyTorch PR
Products & Services
A PyTorch compiler and WebGPU runtime for efficient LLM inference.
A high-performance LLM inference engine in C++ and CUDA, designed as a smaller version of vLLM.
A memory budget solver for PyTorch that reduces peak RAM usage by 20x.
Market Position
Niche focus on low-level ML systems and hardware-efficient inference (WebGPU, CUDA), providing lighter-weight alternatives to massive frameworks.
Leadership
Founders
Jędrzej Maczan
Machine Learning Systems Researcher with a Bachelor's degree in Computer Science from Wrocław University of Science and Technology. He is a regular author for Paged Out! magazine and focuses on the intersection of math, AI, and low-level systems.
Executive Team
Jędrzej Maczan
Lead Researcher
Founder of maczan.pl, AI Researcher focusing on LLM systems. Also holds a day job at DNV applying LLMs to industry solutions.
Founding Story
Jędrzej Maczan started his platform to document his journey into machine learning research, sharing open-source projects and technical insights through his blog and Paged Out! magazine.
Business Model
Revenue Model
Primarily open-source research and contributions, supported by professional employment in the AI industry.
Pricing Tiers
All major research projects and tools are open-sourced on GitHub.
Target Markets
- AI Researchers
- Open Source Community
- ML Engineers
- High-performance AI inference
- ML system optimization
- GPU computing research
- PyTorch
- DNV
- Cohere Labs