02 · Life sciences
Turning scattered study data into an accountable workflow
I designed an AI-assisted authoring system that kept source provenance and human validation visible instead of hiding uncertainty behind generated copy.
01 · Context and ambiguity
The useful problem was underneath the brief.
Study teams assembled critical information across disconnected sources. Faster generation alone would have made the result harder to trust.
02 · Decisions
I made the system explicit before polishing the surface.
- I treated provenance as a product requirement, not metadata.
- I separated source discovery, drafting, review, and approval into explicit states.
- I designed AI suggestions as inspectable proposals rather than silent automation.
03 · Human and AI boundary
Automation got a lane. Judgment kept ownership.
AI found, compared, and drafted from source material. People validated evidence, resolved conflicts, and approved the final record.
04 · Working prototype
The artifact became the research surface.
A working authoring flow connected source evidence, generated content, review status, and unresolved gaps in one model.
05 · Outcome
What changed
The concept gave product and engineering teams a concrete system boundary for an AI feature that had previously been described only as an assistant.
The public case is generalized by domain because the underlying engagement remains under NDA.