Borry
We measure voice, face and body with about 200 open models and the formulas of behavioural science. The LLM does the interpreting.
We bring the signal, not the verdict.
The problem
Today an LLM is given a transcript. People read posture, faces and voices.
These are patterns that science has described, not magic.
A mother sees a bad day at school in a look.
An experienced doctor notices what the patient did not say.
How it works
Record a meeting, a call or a video. About 200 audio and video models extract the raw signals:
Our research engine has read about 2,100 behavioural-science papers and extracted over 2,300 formulas, plus about 500 rules that combine facial action units, each stored with its source and page.
Each one is computed on the recording. Each result is labelled:
An LLM reads all the readings and results and says what happened when. Every claim cites the exact data row it rests on.
video.yolopose.bins#15First result
Same 50-second video, same model, three analyses each, scored by a blind judge.
Read this as a first step. One video, one person; tests on many public videos and on labelled datasets are next. The judge is another call of the same model as the analysts, and there is no ground truth yet.
Demo
A short walk through the product and the test above.
Where it goes
Team
They met in 2016 at Deloitte Italy's Blockchain Lab. Since then they have founded together:
Some are still running; others they have closed or left.