The TREMA lab at the University of New Hampshire works on text-based machine learning, information retrieval, and the evaluation of GenAI and agentic systems. We are especially interested in
We are actively looking to work with industry. We care about real deployments to integrate our research with. We believe that the best questions almost always come from people who are living with these problems day to day. If that is you, here are the ways we can work together.
The first consultation is free. Bring us a concrete problem, for example an evaluation pipeline whose numbers you do not quite trust, a search or RAG system that is hard to measure, or an agent whose quality you cannot pin down. We will tell you honestly what we think, what is known, and whether it is something we would be excited to dig into. No commitment required.
Email Laura Dietz at dietz (at) cs.unh.edu to set up a conversation.
If you are deploying GenAI or agents and your quality dashboard looks suspiciously good, you may be running into circularity: a single-prompt LLM judge that quietly rubber-stamps the very system it is supposed to grade. This is a live focus of the lab, and we work on nugget-based judges and human-AI evaluation workflows that defend against it. See the writeup Which LLM-as-a-Judge Should I (Not) Adopt? for the argument and the references.
This is the kind of problem a membership directly enables us to pursue, in collaboration with the companies who feel the pain first.
Membership is a philanthropic gift to the TREMA lab through the UNH Foundation. It is not a contract for services. In return for their support, members get unique opportunities to engage with the lab, early access to what we are building, and recognition. Gifts fund students, tools, and the open-ended research that makes the rest possible.
Give to the TREMA lab make out donation to “TREMA lab / Dietz”.
All Supporter opportunities, plus:
All Partner opportunities, plus a deeper, individually arranged engagement. This might mean hosting a TREMA intern or student project, co-supervising a thesis on a question you care about, a named seminar, or named recognition for sustained support. Let us talk about what would be most valuable to you.
We can also take on contracted research or consulting when the fit is right. That work is contracted through UNH Sponsored Programs, with the formal agreements, intellectual-property terms, and overhead that come with it.
The simplest first step is an email. Tell us what you are working on and what is hard about evaluating it.
Laura Dietz, Head of the TREMA Lab dietz (at) cs.unh.edu Department of Computer Science, University of New Hampshire