So I built Mimir as a fun side project for myself and my day job. You paste in your raw feedback — interviews, support tickets, survey responses, whatever... and it pulls out themes, ranks them, and tells you what to build next. Every recommendation links back to the actual quotes so you're not just trusting a black box.

There's a real pipeline underneath that does extraction, clustering, synthesis across sources. You can chat with your data, push back on recs, and it learns your product context over time so things get more relevant. When you're ready, it generates specs and GitHub issues so you can hand things off to eng (or, an ai agent).

I've been testing it on public data from YC companies. Obviously some of this is probably way off, but, again it's just using public data so you can imagine it'd be a bitter better with first-party data. And also using it myself....to build it.