Project 22 of 39 · AI systems
Drugs RAG
Retrieval system over scraped drug monographs that answers interaction and prescribing questions by citing the source text rather than generating from memory.
01Overview
A retrieval system built over scraped drug monographs. Ask about an interaction or a dosage and it answers from the source text rather than from model memory.
The scraper had to move from simple requests to a full browser engine, because the source renders content after load and a plain fetch returns an empty shell.
02How it is built
- Answers cite the passage they came from, which is the only defensible way to answer a medical question.
- Handles the awkward parts of the source, including dosage tables that do not follow the page's usual structure.
- The companion scraper produces the corpus this system reads.