Topic
We use language models for the parts of a process that resist rules, such as summarising a CV or drafting a first answer. The model never writes to the record on its own. A person reviews and validates the output, and the pages below show what that looks like in practice.
How we secured Camille, the voice assistant for several medical centers: lightweight local infrastructure, decision models alongside the LLM, and prompts and connections optimized to answer in about one second.
Read moreQwen3.8-Flash-Next locates the optic disc correctly in 120 decisions out of 120, then gets the eye wrong on 51 images out of 60. The protocol, the numbers and the fix we adopted.
Read moreA decision model that answers in a hundred milliseconds is not used the way a language model that answers in ten seconds is. Decision loops, comparing hypotheses, the generator-and-judge pairing, and the return of the small specialized model: the second part of our article on Jev and SemIf.
Read moreFor three years, we have been asking models built to write to make decisions. A new family of models does nothing but decide. First part: what a decision model is, with Jev and SemIf.
Read moreA structured interview tool for hiring: shared evaluation grids, weighted scoring and a candidate ranking your whole team can read and defend.
Read moreA clinic specialised in refractive surgery wanted an honest eligibility simulator. Behind five questions sit a surgeon's real rules, a logic of caution, and total privacy.
Read moreTo hire our junior developer, we built a small web application that brings together the context of the role, the candidate profile, interview note-taking, scoring, and a weighted ranking. Data stays local, code is open on GitHub.
Read moreDiscover the transtorm.ai method for reliable business process automation: an audit of the real process, human validation on important decisions, complete anonymization of data sent to external resources, robust integrations, and continuous improvement.
Read more
