Four databases, one question

Everything a researcher needs to size up a disease already exists. It's just spread across systems that were never designed to talk to each other. Papers are in PubMed. Trials are in ClinicalTrials.gov. Compounds and their clinical stage are in ChEMBL. Protein targets are in UniProt. Each has its own query language, identifiers and gaps.

Live, every time

We query all four at the moment you ask. That costs us some speed, but it means a brief reflects what the sources hold today rather than whenever we last copied them. A full AI brief usually takes under a minute.

Parallel, with a timeout each

The four searches run at the same time, each with its own timeout. If one source is slow or down, the others still report, and the brief says exactly which source failed and why. The AI only ever sees what came back.

This isn't hypothetical. On launch day UniProt stopped answering our checks for a while. Briefs kept working, without protein targets, and said so. Our status page showed the failure in public until UniProt came back.

No sample data

When a source fails, we show the gap. We never fill it with example records. A research tool that invents data to look complete is worse than one that admits what it doesn't know.