← All work

Nonprofit · biomedical research funding

Human-in-the-loop AI for executive thought leadership

An assistant that watches scientific journals and newsrooms, scores items for relevance, and drafts LinkedIn comments in an executive's voice. She reviews every draft in Teams and posts it herself.

  1. Journals & news
  2. Relevance scoring
  3. Drafts in her voice
  4. Teams review
Human-in-the-loop AI for executive thought leadership
IllustrativeIllustrative: a drafted comment waiting for review.
Client
Alzheimer's Drug Discovery Foundation ↗
Nonprofit · biomedical research funding
Stack
Python 3.12, FastAPI, Postgres, Gemini, AWS Lambda, EventBridge, Power Automate, Microsoft Teams

Results

  • Never logs in to, scrapes or posts to LinkedIn
  • 62 unit tests, with eval gates before prompts change

What I built

  • Monitoring with de-duplication. A low-cost model scores relevance and a stronger model drafts, and every number in a draft is checked against its source.
  • Delivery to a private Teams channel as Adaptive Cards, plus a paste-a-post path for posts she finds herself.
  • Learning from her edits. Her edits tune the voice, with a human review step and eval gates before any prompt changes.