Assistance without a language model

Fixed rules point. A person decides.

ValidTrials has a rule-based assistant. It points at gaps and inconsistencies with the records behind each finding. It does not use machine learning or a language model.

What the assistant does

Study review

Fixed rules check the study set-up, such as overlapping visit windows or endpoints without an analysis. Each finding lists the records it is based on.

Method suggestions

A lookup table suggests standard methods for an outcome type and design. The statistician chooses the method.

Assumption checks

For a finished workbench run, the assistant lists the assumptions of that method and what the output shows about them.

Unusual values

Simple statistical rules flag outlying values in the case-report data for a data manager to review.

Document suggestions

Sections of a study document can be pre-filled from records already in the system. A person accepts each suggestion, and nothing is approved automatically.

What it does not do

No language model

No text is generated by a language model, and no study data is sent to an AI provider.

No automatic coding

MedDRA and WHODrug terms are typed by a person who holds the licence. The system does not propose codes.

No writes to the record

Findings are advice. The assistant does not change clinical data, open queries, or sign anything.

No AI-system endpoints

ValidTrials does not run AI models as part of a study. For AI-enabled devices, the algorithm version used per participant is recorded.

Bring us the protocol.

We read the study, configure ValidTrials, and the sites work in one record. The first conversation is about your protocol, not a software tour.

Further reading