Aayush Garg

ML engineer who ships production AI systems and rebuilds LLMs from first principles.

I am a Senior ML Engineer specializing in computer vision and LLMs. At Jiffy, I build the AI image enhancement pipeline powering a $100M+ DTF printing business and Smart Redraw, a multi-agent image editing system. Outside work, I build the full LLM stack from first principles and write about it here.

💼 Current Work

I work at the intersection of computer vision and LLMs, building production AI systems. On the side, I am building the full LLM stack from scratch: BPE tokenizers, GPT-2 pretraining, SFT and GRPO alignment. My background in research and computational geophysics keeps me drawn to understanding things from first principles.

If you are interested in my work or want to collaborate, feel free to reach out via email or connect on LinkedIn or Twitter.

📮 Blog

I write about LLMs, concepts deep-dives and practical AI engineering. Below are my recent posts.

Date	Title
3/29/26	Using a local LLM in OpenCode with llama.cpp
3/27/26	FlashAttention: Making Attention I/O-Aware
3/20/26	Building Browser Tools with Claude Code
2/26/26	GRPO: Building Intuition Through Ablation Studies
2/12/26	Tools for Better Claude Code and Terminal Experience
1/23/26	Expert Iteration for Math Reasoning
1/13/26	A Brief Introduction to Claude Agent Skills
1/6/26	Key Insights from DeepSeekMath paper
1/1/26	Understanding GRPO: PPO without the Critic
12/30/25	Deriving the DPO Loss from First Principles

📬 Reach Out

The best way to follow my work is on GitHub, Twitter, or LinkedIn.

Template inspired by hamel.dev and Lil’Log.