Rebekka Mikkola, Co-founder & CEO, Trismik | Slush 2025
summary
Trismic adapts computer-adaptive testing to evaluate LLMs, running MMLU Pro on Llama 3.2 in 26 seconds versus 25+ minutes for classical tests, a reported 98% time saving and up to 90% cost reduction by cutting unnecessary runs. The platform optimizes sampling and uncertainty, supports adaptive, classical and stratified tests, shows consistent accuracy, offers a beta with dataset library and upcoming user data upload in Q1, and has a $3M pre-seed round and academic partners validating results.
AI-written summary · about
related talks
38mMaking LLMs reliable - A practical framework
43mEval-Driven Development: Mastering Agentic Tracing and Expert-Aligned Judges With MLf...
35mEvaluating LLM systems: the nuts and bolts
17mLLM Observability, Evaluation, Experimentation Platform
31mDeploying Large Language Models with Ease
37mToward More Efficient Triage With AI & Data- 17mLLM Observability, Evaluation, Experimentation Platform
28mTrustworthy LLMs: Vibe checks are not all you need