Trismik Secures £2.2 Million Pre-Seed Funding

Trismik, a science-grade LLM evaluation startup, led by Rebekka Mikkola, Nigel Collier, and Marco Basaldella, has raised £2.2m in pre-seed financing led by Twinpath Ventures, with participation from Cambridge Enterprise Ventures, Parkwalk Advisors, Fund F, Vento Ventures, and angel investors from Ventures Together.

As AI labs race to build more powerful models, a key challenge has emerged: measuring what these systems can actually do. Traditional benchmarks are saturated, multiple models now score above 90% on tests like MMLU and GSM8K, making it difficult for businesses to evaluate task performance and communicate results. Public benchmarks often fail to reflect enterprise-specific requirements or proprietary data, while new regulations mandate specific assessments for compliance. Teams face a perfect storm: they lack granular performance metrics and adaptive testing capabilities, even as competitive pressures demand faster development cycles.

A team from the University of Cambridge, now operating under Trismik, is exploring a new approach to evaluating next-generation AI models. They propose combining proven psychometric methods for human intelligence with adaptive tests that adjust difficulty in real time.

Trismik’s team, which includes former engineering and commercial leaders from Amazon and Salesforce alongside Cambridge scientists, applies Item Response Theory (IRT) and Computerized Adaptive Testing (CAT), the scientific basis of standardized human intelligence tests, to AI evaluation.

Collier, who began his career in the 1990s with a PhD in neural network–based machine translation, has focused on ensuring AI serves as a trusted partner rather than a risk. A prolific researcher with over 200 papers, his curiosity about assessing AI as efficiently and fairly as humans inspired Trismik’s proprietary adaptive evaluation approach.

In 2023, Collier met co-founder Rebekka Mikkola, a repeat founder and enterprise sales executive passionate about AI and promoting women in tech. Backed by Cambridge Enterprise and partnered with a major UK telco, they built Trismik’s first MVP. In 2025, former Amazon scientist and TEDx speaker Marco Basaldella joined as CTO, completing a founding team combining deep scientific, engineering, and commercial expertise.

Trismik’s approach could transform how model capabilities are measured, shifting from crude accuracy percentages to nuanced ability distributions. For organizations spending $100,000+ monthly on GPU compute, adaptive testing could cut evaluation costs by up to 95% while delivering near-identical rankings, Spearman correlations exceed 0.96 on 4 of 5 datasets using just 8.5% of the questions.

Following its funding, Trismik will launch its product for AI developers, offering classical and adaptive evaluation across popular public datasets covering factuality, alignment, safety, reasoning, and domain-specific knowledge, giving scientists broad, actionable insights into model performance.

Trismik’s LLM Experimentation Platform currently offers lightweight, accessible tools for fast, reliable evaluation. The company plans to expand it into a full environment for designing, running, and analyzing LLM experiments, including fine-tuning and prompt engineering. Early access will be available via Trismik’s website, with onboarding beginning in late 2025.