A pharmacovigilance agent that answers drug safety questions
with figures you can trace back to the source.
Drug safety teams answer the same question all day: is this adverse event a known effect, a coincidence, or a signal? The data is public but scattered across EudraVigilance, FAERS and the MHRA Yellow Card scheme, each with its own format, and the reference (section 4.8 of the product label) lives in PDFs. In this field, a chatbot that returns a number nobody can verify is worse than no answer.
An agent built at Kirha. It works out what is being asked (drug, effect, population, comparator), collects case counts from the three regulators, reads the label's adverse reaction table with its frequency bands, and matches reported reactions against labelled ones. It computes disproportionality (ROR, PRR) under strict guardrails: a frozen denominator, no mixing across databases, and a plain reminder that disproportionality is not causality. Every figure ships with its inputs and its source link, so a pharmacovigilant can redo the calculation. Case files dropped in by the user are read through openicsr.