Taking new projects Scope <24h Replies within a few hours
Tesryn

Tesryn Lab · AI evaluation

Research Runtime

An AI researcher that has to prove it's better than the last one before it ships.

In the workshop For teams that need answers with sources behind them · 2026

In the workshop. Designed and agreed, not started. If you need it, it moves up the list.

The assistant our benchmark was built to measure: it reads the documents you point it at, answers the question you actually asked, and shows the passage behind every claim. Nothing ships until it beats the current version on a test it has never been tuned against.

The moment it will earn its keepA question that would have cost an analyst a day comes back in minutes, with every claim attached to the document it came from.

What it proves we can do

Four things it will do.

01

Answers with their sources attached

Every claim carries the passage it came from, or it doesn't get made.

02

Measured before it ships

Each version has to beat the last on a sealed test.

03

Works on your documents

Point it at what your team already has, wherever that is.

04

Runs where your material can live

Sensitive material can stay on hardware you control.

What it shares with the rest of the Lab.

The same things we know how to do keep coming back in different industries. This is where this one sits: what it proves, and what else proves the same thing.

Every claim traced to its source Nothing is asserted without the passage, part number or measurement behind it. Also proved by Podcast Maker, Engineering Library, Reconstruction Research Studio, Part Finder, Application Engineering, Research Benchmark.

Knowing whether a change actually helped Conditions written before the run, a sealed test, and a verdict that stands either way. Also proved by Research Benchmark, Capability Distillation.

Runs on hardware you control Models and data that never have to leave the building when the material is sensitive. Also proved by Podcast Maker, Engineering Library, Command Center, Sentinel.

The whole map — six sections, and everything that proves them.

  • No version ships on a feeling. It ships on a measured result.
  • An answer without a source is a bug, not a style.

What it isn't. Not built yet. The instrument that will judge it has been running for months.

We'll tell you what it will do. Not how it's built.

Yours would be your own build, on your systems, and you'd own it outright: code, documentation, no lock-in. What we bring to it is the practice — we've solved this shape of problem before, so you're not paying for the first attempt. The method stays ours. And if what you need isn't here at all, that's the usual case — it's how every skill on this page started.