Sourced
Sourced is a provenance-first medication safety review tool for clinicians and pharmacists.
Core trick
Retrieves interactions only from cited FDA label, DDInter, and FAERS data, and a verifier agent rejects any claim beyond its quoted source.
Description
Sourced is a provenance-first medication safety review tool for clinicians and pharmacists. It ingests a patient's context (clinical note, medications, allergies, diagnoses, labs) and returns a ranked, source-backed review of medication interaction risks. Its core rule: it never asks the model for a clinical fact. Every interaction, severity, and mechanism is retrieved from cited sources (FDA drug labels, DDInter, openFDA FAERS, RxNorm), and an adversarial reviewer agent rejects any claim that goes beyond its quoted source, so it is never shown. It handles Spanish and English medication lists, and when a drug pair has no graded severity in the database, it does not guess: it names the real mechanism it found and routes the question to Claude Science for deeper research. This matters because medication harm is nearly half of all preventable harm in care, and transitions like discharge are where it happens. Unlike a pairwise checker, Sourced contextualizes each risk to this specific patient. Link: https://sourced-lake.vercel.app
README excerpt
# Sourced Medication safety review, backed by cited sources, never by model recall. [](https://github.com/Railly/sourced/actions/workflows/ci.yml) ## What it is Sourced is a medication-safety review tool for clinicians and pharmacists. Give it a de-identified patient context (clinical note, meds, allergies, diagnoses, labs) and it returns a ranked, source-backed review of medication risks. It does not diagnose or prescribe. ## The one idea We never ask the model for a clinical fact. Every interaction, severity, and mechanism in a Sourced report is derived from cited sources and re-checked. Anything that cannot be traced back to a source is refused, not guessed. A second, adversarial verifier agent runs after synthesis and strips out any claim it cannot trace to a source object. When Sourced hits the edge of what it can resolve from cited sources (a drug pair the database documents without a graded severity, or a flagged concern it cannot trace) it does not just drop it. It generates a precise research brief and routes it to Claude Science, Anthropic's AI workbench for scientists, opening a real research session backed by 60+ scientific databases. Sourced finds the boundary of its own knowledge. Claude Science crosses it. Refuse to assert is defensive. Refuse, then route to research, is generative. ## Data layer - **DDInter 2.0** — 222,383 interaction rows, 160,235 unique pairs, 1,939 drugs.