Applied AI / Research pipeline
In daily useCross-review research before it becomes a draft.
Independent model passes, rotated critiques and a separate audit of the final draft. A human writes the brief and decides what gets published.
Why I built it
A plausible draft can quietly turn uncertainty into fact. I built a workflow that keeps independent explorations and critiques available when reviewing the final text.
Engineering decisions
Start with five models exploring the same human-written brief independently.
Rotate cross-review before synthesis, then use a separate writer and an auditor from another model vendor.
Check the draft against the research outputs and return it to a person for the publishing decision.
What works today
The project notes record 60 topics and 255 waves since March 2026. Outputs have informed project research, pitch materials and essays on this site.
Current boundaries
Model disagreement is a review signal. It does not establish factual correctness. The code is private and the workflow never publishes autonomously.