Trismik

Trismik is a University of Cambridge spinout aiming to set a new standard in Large Language Model (LLM) evaluation. Incorporated in January 2025, the company is developing an adaptive testing platform to address critical bottlenecks in AI safety and reliability. The founding team combines deep academic research with commercial experience. CEO Rebekka Mikkola previously worked at Salesforce, Chief Scientist Nigel Collier is a Professor of Natural Language Processing at the University of Cambridge, and CTO Marco Basaldella brings experience from Amazon where he focused on foundation models. The venture originated from Professor Collier's research at the university's Language Technology Lab, which identified the scalability limitations of traditional LLM testing methods.

The company's core business is providing AI engineers and developers with tools to rigorously test LLMs for risks such as hallucinations, bias, and factual inaccuracies more efficiently. Trismik operates in the rapidly growing LLM and Generative AI market, serving enterprise partners in sectors like media, telecom, and technology. The business model centers on its evaluation platform, which uses adaptive testing and automatic scoring to assess AI models. This approach, inspired by psychometrics, dynamically selects the most relevant test questions based on the model's performance, significantly reducing the number of queries needed compared to static benchmarks. This allows for much faster and less costly development cycles. For example, Trismik demonstrated a 98% reduction in evaluation time for the Llama 3.2B model.

Trismik's platform is designed to make LLM evaluation faster and more trustworthy, delivering results up to 180x quicker than conventional methods. Its key technology is Computer Adaptive Testing (CAT), which tailors the difficulty of test items in real-time based on the model's responses. This allows the platform to identify model weaknesses and failure modes that standard benchmarks might miss, using fewer than 80 queries in some cases. The platform is being developed to include a developer-friendly SDK and a comprehensive test library. After participating in the Founders at the University of Cambridge START 2.0 accelerator and securing a Technology Investment Fund award in 2024, Trismik has raised pre-seed funding and is engaging in commercial pilots. A full product launch is planned for September 2025.

Keywords: LLM evaluation, adaptive testing, AI safety, model evaluation, generative AI testing, AI risk management, trustworthy AI, bias detection, hallucination detection, AI testing platform, natural language processing, Cambridge University spinout, psychometric testing, automated scoring, MLOps, LLMOps, AI engineering tools, developer tools, enterprise AI solutions, AI model validation