Connecting user signals
to action with AI agents.
Feedback arrived through forms, Slack, email and the platform. I built connected AI workflows to bring repeated reports, usage signals and prior decisions together before deciding what to change.
Key decisions
- Separated triage, investigation and refinement into three agents, each with a defined handoff.
- Checked reports against internal records, usage signals and previous decisions before preparing revisions.
- Built subject-specific Claude skills: the Economics pipeline checked theory accuracy, while the Geography pipeline checked information daily to keep it current.
- Kept evidence assessment, fact-checking and approval with me.
Results
- Reviewed all 429 topic articles across Economics (158) and Geography (271) over two months, with no factual corrections needed after review.
- Connected grouped feedback, supporting evidence and revised drafts in one review process.
- Published guides recorded 4,400+ student reading hours in 30 days. This engagement figure does not isolate the effect of the agents.
Reflection
An individual report rarely supplied enough context for a decision. I needed to check it against usage and previous work before acting.
My contribution
- Connected feedback sources to a scheduled agent that filters, groups and deduplicates reports into a reviewable draft.
- Prompted searches of internal records and feature-management systems, using SQL where needed. For feature ideas, included competitor research and prior decisions.
- Built Claude skills to guide structure and style, and generated images to brand requirements.
- Reviewed the exact passages flagged by audits, checked the claims and guided the next revision.
Forms · Slack · Email · Platform feedback
My review guides the next revision.


