[mustaphah]

Well, that's interesting.

I doubt it's feasible without human curation (of the content produced) in the loop - at least not with the current state of AI. Also, writing docs should be a team practice. A tool like this could actually cause more harm than good if engineers lean on it too heavily. I mean, interviews capture a moment in time; without regular updates, context will drift.

More importantly, this leans hard on human memory, and most people have terrible ones. That can be partially fixed by prompting to explain why they made specific tradeoffs proactively. For example, the LLM could detect that a tradeoff was made and then ask: why this one? And potentially list the alternatives.

It's a cool idea. You can build a proof-of-concept in a few days with AI prototyping tools.