How it works
The content team adds a topic in Airtable, a Make.com workflow retrieves relevant context from a Pinecone index of the charity's own library through a Cloudflare Worker, and a draft in their voice is generated and routed to human review. A scheduled hash check re-ingests any source page that changes.
What triggers it
A new topic row in Airtable.
Step by step
- 1The library is ingested with Apify and OpenAI embeddings into Pinecone
- 2A topic added in Airtable starts a Make.com workflow
- 3A Cloudflare Worker retrieves the most relevant passages
- 4A draft grounded in those passages is written and sent for human review
- 5Changed source pages are detected by MD5 hash and re-ingested
Why an off the shelf connector was not enough
Generic AI writing tools cannot be limited to an approved medical library. Grounding every draft in the charity's own sources needed a retrieval layer built around their content.