Skip to the content
ΕλληνικάTalk to us

techmirion.com

R&D · Research

What we study and why.

We build private AI for work that has to stand up to checking — a measurement, a deadline, a number someone has to sign off. The question we study is how a system runs the same way every time, on your own machines, so that every answer can be checked: same question, same answer, and every result shows where it came from. Our products grow out of this research, and when the two pull in different directions, the research wins.

01Our first paper

Adamidis et al., 2026 — May 2026

One real measurement from the lab, a photoluminescence analysis, went to our tool and to three leading AI platforms, all on paid plans with every feature turned on. Each one got the same file and the same plain-English request, four times. Our tool gave the same result in all four runs. The three commercial platforms either varied from run to run, in their numbers and in how they analysed it, or produced no usable result on the task. The point is in the paper's title: the consistency comes from the tool, not the model.

02Done since the paper

  • a re-test on today's AI tools with many more runs (17–18 October 2026)
  • a second instrument (an electron-microscope pattern)
  • a privacy test on data that may not leave (three invented files)
  • thirty harder questions on which the private model scored level with the top paid models

The full runs are on the proof page

03What we're working on next

  • a formal study with users
  • support for more lab instruments
  • systems at more organisations

04Next: systems that improve their own method.

On our own hardware, our model now proposes rewrites of its own tool descriptions and instructions. Every proposal is checked for false facts, tested blind against the unchanged version on questions it has never seen, and kept only if it wins by a clear margin. A kept change is then tested again on a sealed set of questions that nobody tunes on.

The first rounds, in October 2026, kept some changes and rejected others. We are re-testing the kept ones on larger samples before we print a figure.

What has to happen before a change reaches anyone

  1. 1. Fact check
  2. 2. Blind test against the unchanged version
  3. 3. Re-test on a sealed set
  4. 4. A person approves

We think private AI goes this way next: a model on your own machines that gets measurably better at your work, under your control.

05More

Start with one process.

Talk to usOr write to [email protected] (subject: Research)

Sources