Research, writing and books, in one place
Read it. Move it.
Keep it.
Projects with their articles and papers, essays in series, and books where every number moves. Everything is open to read.
How reading works here
- Read anything, signed out. Nothing you read is kept; a page is only counted, with no cookie.
- Sign in and your place is kept, with five questions on what you read.
- What you finish comes back for recall: in 1, 2, 5, 11 and 24 days.
hanumemAI proprietary
Long-term memory for an assistant. The highest published score on LoCoMo, 96.1, reading 1,533 tokens per question where the systems beside it read 6,700 to 45,800.
HanuDB open source
A document database for one small machine whose durable home is a bucket. Delete the machine and a new one rebuilds from the bucket alone.
Documents that cite themselves proprietary
Reads a form or a scan, points at the exact words for every value, and says which values a person must check. 97.53 ANLS on all 5,349 DocVQA validation questions.
HanuAnalytics open source
Product analytics and session replay on one free machine, with a bucket as its home. The machine is free, the storage is cents, and the data never leaves your project.
Think Like Me proprietary
A short voice interview becomes a twin that answers the way that person would, in their voice. Scored on questions the interview never asked.
Speak As Me proprietary
A short recording of a person becomes a voice that answers in their words while they are away. From a clean clip the voice is as close to the person as their own other recordings; from a poor one it gets about four fifths of the way.
Split-second decisions proprietary
A voice agent makes several small decisions in every turn of a call. Handed to a small decision model, about 39 in 40 came out right, each in under a quarter of the time allowed.
BigKids proprietary
A maths and English tutor for young children that reasons before it drills. In simulation, its lesson planner now gives a child a lesson they are not ready for about three quarters less often.
Measurements
The four metrics that tell you whether a RAG or LLM system actually works — each one explained on its own.
seriesRAG
Retrieval-Augmented Generation, built naive and improved one measured step at a time — Novice to Ninja.
seriesModels in the browser
Gemma 4 running inside a browser tab on WebGPU: no server, no API key, no credit wall — and what that does to the frontend job.
seriesBuild notes
Short write-ups of things I built, most of them in a weekend.
How an LLM Works: Novice to Ninja
What happens between your question and the answer: tokens, weights, one multiplication repeated billions of times, and a guess. Taught from zero, with a real model opened up on the page.
interactiveAI Agent Memory: Novice to Ninja
How an AI assistant remembers you, taught from zero with one card box, one secretary, one librarian, and numbers you can move with your own hands.
interactiveAI Agent Loops and Graphs: Novice to Ninja
Loop engineering and graph engineering for people who have used agents and want to build them: loops first, then graphs, then real problems. One small open-source library, agl, is the companion code.
interactiveAutoresearch: Novice to Ninja
Karpathy's overnight research loop, taken apart: one file the agent edits, one number nobody edits, a five-minute budget, keep or discard, and a program the human improves between nights. With the reported results, five real ledgers of our own, and a loop you can run on a laptop.
interactiveDesign a Real-Time Voice AI Agent: Novice to Ninja
A system that listens, thinks, acts and talks back within a fraction of a second. The five components, the two ways to build it, the 800 millisecond budget, interruptions, tools, telephony, scale, the edge cases, and how to present it all in an interview.
interactiveDecision Models with Jev: Novice to Ninja
Choice, Noul and Score from the first call to a production design: how a decision model stamps typed answers with probabilities, and how your own code does everything else. Every claim is measured on one running example, a building's maintenance desk.
bookThinking in Vectors
Meaning as geometry: what a vector is, where embeddings come from, how search over millions of them works, and what it takes to run it.
interactiveNotes: One Page Each
Short notes on one thing worth knowing. Each is a single page with its sources, read in the time it takes to make tea.
interactiveMake an Interactive Book: In One Afternoon
How the books on this shelf are written: a folder of plain HTML chapters, a few sliders, and one rule about numbers. Copy this book and replace its words with yours.