
Processing 5GB audio files: building a resilient media pipeline
Why we built an audio splitting pipeline, what broke when real users started uploading browser recordings, and the fallback layers that made the system resilient.
Ce que nous construisons et comment nous pensons. Mises à jour produit, plongées techniques et leçons tirées de la création d'une plateforme IA pour les comptes rendus de réunions.

Why we built an audio splitting pipeline, what broke when real users started uploading browser recordings, and the fallback layers that made the system resilient.

From first commit to streaming AI chat, 671 commits later. The decisions that paid off, the ones I'd make differently, and what I learned building a full-stack AI product as a solo founder.

The real productivity bottleneck is not the meeting. It is the 90 minutes afterward, translating decisions into deliverables. Most meeting AI focuses on the wrong phase.

Most meeting tools stop at telling you what was said. The real value is generating the deliverable you would have written manually: the spec, the brief, the email, the backlog.

Independent consultants spend 25% of their week on post-meeting write-ups. That is where margins shrink. There is a faster way.

The thinking was done in minutes. The document took hours. That translation step is where momentum dies, and it does not have to be manual.