Today I'm excited to announce that we are launching @hyperparamapp, an AI-powered Swiss Army knife for massive LLM datasets. It lets you view, score, filter, label, and transform LLM data directly in the browser. I started Hyperparam one year ago because I knew that the world of

Someone leaked our architecture diagram 😲 What if our competitors learn that you don’t actually need a backend??

Agree! The optimal amount of risk is not zero.

The human in the loop interface is how the user expresses what they want from their AI. And the human-in-the-loop is what distinguishes high-quality, well-engineered software from slop. It’s something we believe strongly in at Hyperparam.

https://t.co/XENU4GsfcQ

So much interesting data in the open router state of ai report! There's a chart for everyone in there

There’s recently been discourse about AI sentiment on X vs Bluesky. A few weeks ago I spoke in-person with a progressive political group about AI. I'm being honest, I knew the moment I was invited that this was going to be a controversial topic for the liberal crowd. But I really

In the last few weeks I’ve shipped a couple new projects that would have taken FAR longer without AI. I singlehandedly wrote an entire SQL engine from scratch with over 93% test coverage in less than a week. You can judge for yourself if it’s slop or not. https://t.co/kFKL3Si71b

AI != Slop I’ve been writing code professionally for over 20 years, but today, I barely write any code without the use of AI. I use AI to produce more and BETTER code than I could without AI. And that takes work. You need creativity, discipline, and an understanding of what

three years now since the world changed

I’m in between gigs and can actually be productive, so I decided to spy on Claude Code. I built Snoopty - a simple proxy and UI for snooping on Claude Code. You start the server and proceed as normal. But now you’re the spy master and can 👀 all messages.

When someone says they created a project in record time thanks to AI, I've seen snarky comments like "github or it didn't happen". Well... judge for yourself. https://t.co/kFKL3ShzbD

I created and released a new JS library this week: Squirreling SQL Engine Existing in-browser engines like duckdb-wasm don't handle async well. Squirreling JS is built to query over async data sources (like parquet), minimize fetches, and stream data back as an AsyncGenerator.

If you come across a Parquet dataset in your work, what do you do? Download it and load it in Jupyter? Why do we still accept this multi-step annoying process as normal while it’s been proven that all you need is the browser?

https://t.co/NkHMYFBVVR

The amount of text being produced by models is overwhelming, and only increasing. On Wednesday, we’re launching Hyperparam, the first tool that transforms LLM-scale datasets like it’s nothing: - Drag and drop your dataset - Instantly explore with natural language - Score,

This is something I feel strongly about. I hate how backend is considered more "real engineering" than frontend. How many companies have you been at where product is developed "backend first" and then APIs thrown over the wall to the frontend team and told "make me a pretty UI".

When @huggingface adopted my open source libraries, I realized there was something more to what I was building. In this Q&A I talk about my lessons learned from a year of Open Source software: https://t.co/wCyWd08mFn

I was frustrated with the experience of trying to work with chat logs in jupyter notebook. There has to be a better way. That’s the mission I’ve been on for the last year building Hyperparam.

How do you avoid abstraction? I look at the whole system and what it's goal is (eg- view data in the browser), and then ask what would be the minimal code required to achieve that goal if you included the ENTIRE stack (including frameworks, libraries, infra, etc). I think this