Cambridge University spin-out Trismik is building an AI evaluation platform that borrows Item Response Theory and Computerised Adaptive Testing from human IQ testing to measure LLM capabilities. The system dynamically adjusts question difficulty in real time, delivering near-identical accuracy rankings while using only 8.5 % of the test items, which the company says can cut evaluation costs by up to 95 %. Current features cover factuality, alignment, reasoning, safety and domain knowledge, and early access has already been validated on seven models across five benchmark datasets. The capital will fund product launch and expansion, with enterprise onboarding targeted for late 2025 and a full solution slated for early 2026. Longer-term, Trismik envisions the platform evolving into a broader experimentation environment incorporating fine-tuning, prompt engineering, compliance tracking and performance visualisation. The team combines academic depth—over 200 NLP papers from Chief Scientific Officer Prof. Nigel Collier—with commercial experience from CEO Rebekka Mikkola and CTO Marco Basaldella. No revenue or user numbers have been disclosed, indicating the company remains in its formative, pre-revenue stage.

Total raised

$3M

Funding rounds

1

Latest round

Seed

Latest activity

Sep 2025