Money is moving on AI claims.

Technical due diligence and defensibility assessment of AI and machine-learning products: whether the model does what the deck says, whether the moat survives the next generation of foundation models, and whether the data advantage compounds or evaporates. The findings come back in the documents deal teams already use, which is a red flag report when you need a fast go/no-go and a full DD memo once the deal advances.

Where a claim needs testing rather than reading, Datatroph builds the evaluation harness, and it stays with you after the engagement.

Datatroph has been moving data and making it useful since 2008, across a wide range of industries. The AI work is recent and now accounts for nearly all of it, with a good deal on clinical and genetic data. Recent engagements include cancer-detection and therapy-target products, clinical scribing companies, AI products with no connection to medicine, and the buy side itself as a Venture Partner.

Brian Hur, who runs Datatroph, co-founded VetPronto and ran it as CEO through YC W15, and later ran the technology budget as chief information officer of a veterinary group in Australia; both roles are in the record below.

Datatroph partners with researchers to use AI inside their projects: deciding where a model can do real work, such as labelling free-text records, screening the literature, or linking clinical data to other sources, and where it would only add noise; building the pipeline alongside the research team rather than handing over a black box; and setting up the evaluation so the results hold up under peer review.

Brian Hur has done this work from the research side, from a PhD that taught language models to read veterinary clinical records for epidemiologists to more than twenty peer-reviewed papers with coauthors in veterinary medicine, epidemiology, and NLP, including a guide for veterinarians on how to evaluate AI tools.