Top all-time · #22309
Henry Pan
Personal blog · henrypan.com/blogaiprogramming
Henry Pan writes Henry's Blog, a personal blog at henrypan.com about ai and programming. Since 2025, 3 different Hacker News readers have linked it 17 times, most often “I Built a Deep Learning Library from Scratch Using Only Python, NumPy and Math”. It is still writing: the last post was on 2026-07-19, and you can follow it by feed.
Henry Pan is a machine learning engineer working on LLM post-training and training optimizations, with experience in recommendation systems and scalable ML infrastructure.
Most-linked posts?
- I Built a Deep Learning Library from Scratch Using Only Python, NumPy and Math2 people · 5 mentions · first linked 2025-02-09
- Harness Training2 people · 4 mentions · first linked 2026-07-22
What Hacker News says?
I built a machine learning library [1] (similar to PyTorch's API) entirely from scratch using only Python and NumPy. It was inspired by Andrej Karpathy's Micrograd project [2]. I slowly added more functionality and evolved it into a fully functional ML library that can build and train classical CNNs [3] to even a toy GPT-2 [4]. I wanted to understand how models learn, like literally bridging the gap between mathemat…
Comment linking /blog/2025-02-06-ml-by-hand/
on “Ask HN: What are you working on? (March 2025)”Submitted Harness Training as a story · 3 points
HN discussionI'm continuing to expand my own deep learning library [1] built with numpy-primitives to support LLM post-training techniques like supervised fine-tuning (SFT) and reinforcement learning with GRPO. It's a good learning experience to work without all the high-level abstractions to "build a wheel" and "use that wheel to build a car". I'm also looking into coding harness self-improvement [2]. An inner LLM (raw LLM requ…
Comment linking /blog/2026-05-25-self-improvement-harness/
on “Ask HN: What are you working on? (June 2026)”Submitted I Built a Deep Learning Library from Scratch Using Only Python, NumPy and Math as a story · 3 points
HN discussionI did something similar. It started off as a "self-improving agent" project, inspired by autoresearch, then later on I reframed it as "harness training" (discrete program search) borrowing the mental model from ML training. I "trained" the harness on a subset of Terminal-Bench 2.0 tasks while keeping the LLM (local Qwen3.6-35B A3B) frozen. Making LLM inference and the task environments fully deterministic was necess…
Comment linking /blog/2026-07-18-harness-training/
on “Harness Engineering for Self-Improvement”Submitted Training Agent Harness Like Training a ML Model as a story · 2 points
HN discussionI recently wanted to see whether an AI agent could self-improve a harness to solve terminal bench tasks. It’s possible for an AI agent to propose a meaningful one-time change to the harness, but after experimenting with this for a couple of weeks, I think the continuous self-improvement is mostly an experiment-systems problem. The system needs a way to decide what kind of improvements can safely compound. Turns out…
Comment linking /blog/2026-05-25-self-improvement-harness/
on “What 1k Harness Experiments Taught Me About Self-Improving Agents”Submitted What 1k Harness Experiments Taught Me About Self-Improving Agents as a story · 2 points
HN discussionI worked on this project ( https://github.com/workofart/harness-training ) for the past few months to reframe "Agent-driven Self-improving Harness" to "Harness Training". The idea is simple, the harness is trained once with a frozen task LLM against a given task environment. Then you can then swap out the task LLM to any model and evaluate the "frozen trained harness" with any task LLM on any new task environment. S…
Comment linking /blog/2026-07-18-harness-training
on “Ask HN: What Are You Working On? (July 2026)”Submitted I Built a Deep Learning Library from Scratch Using Only Python, NumPy and Math as a story · 2 points
HN discussionThanks for sharing! It's inspiring to see more people "reinventing for insight" in the age of AI. This reminds me of my similar previous project a year ago when I built an entire PyTorch-style machine learning library [1] from scratch, using nothing but Python and NumPy. I started with a tiny autograd engine, then gradually created layer modules, optimizers, data loaders etc... I simply wanted to learn machine learn…
Comment linking /blog/2025-02-06-ml-by-hand/
on “Build a Deep Learning Library”Submitted Freeze the model, train the harness: gains transfer across LLMs and benchmarks as a story · 1 points
HN discussionSubmitted Self-Improving Harness Is an Experiment Design Problem as a story · 1 points
HN discussionSubmitted Show HN: What 1k Harness Experiments Taught Me About Self-Improving Agents as a story · 1 points
HN discussionSubmitted How a Deep Learning Library Enables Learning as a story · 1 points
HN discussionSubmitted How A Deep Learning Library Enables Learning as a story · 1 points
HN discussionSubmitted Understanding ML: Creating a PyTorch-Inspired Deep Learning Library from Scratch as a story · 1 points
HN discussion