Ndovu elephant mark ndovu

NDOVU · AI RESEARCH LAB

Inside the lab.

Deep dives on AI, Agentic Engineering, and Security - hands-on experiments, audio explainers, and a podcast, all published in the open.

Distill Podcast 2026-11-16 All episodes →

The Model Took the Lead. Who Checks the Work?

UC Santa Barbara physicist Nathaniel Craig says GPT-6 Astra took the lead on the detailed proof in his preprint proposing a candidate lattice definition of the…
0:00 / 11:35

Distill is the Ndovu lab podcast - AI on Mondays, Hacker News on Wednesdays. Made for your walk.

Lab-sites

3

Standing, evergreen platforms - each a self-contained site on one topic: consoles, specs, explainers. They grow for years.

Research

3

Evidence campaigns and deep dives - what we ran, what held up.

2026-11-01

AI Code Review: What the Evidence Shows

Six parts of the AI code review practice, compared across 18 engineering teams and 195 evidence rows: the noise war, the merge gate, cost, the reviewer's own attack surface, vendor data flow, and what nobody has measured. Every number carries its source; the retracted claims are published too.
research
2026-09-20

How useful is Jev, really?

Seven experiments, five models, identical harnesses: we ran the System One model through real lab workloads - HN classification, Kenyan business news, live triage, browser agents, robustness probes, guardrails, and code judgment - and measured what holds up.
research
2026-09-18

Jev & System One Models

TypeSafe AI calls Jev the first System One Model: typed, probabilistic decisions in milliseconds at a token price that looks like a typo. What holds up, what does not, and when to reach for it.
research
0:00