Problem Statement

Status: Accepted

Author: Ryan Nel

Date: 2026-04-18

---

Observed Problem

Users need a way to capture meetings — both in-person and virtual — directly from the WordLoop app without relying on third-party recording tools or manual note-taking. Today, meetings can only enter the system via file upload or manual text entry — both of which are post-hoc and require the meeting to have already happened. There is no live capture flow.

The core pain points are:

1. No live capture — users must remember to take notes or record externally, then import later.
2. No real-time feedback — users have no visibility into what's being captured until processing completes.
3. Post-processing delay — insights (talking points, tasks, topics) are only available after a batch transcription and synthesis pipeline finishes.

---

Appetite

Large. This is the most complex bet the platform has run so far. It introduces a binary audio streaming path, a bidirectional HTTP stream between Core and ML, a new recording lifecycle, speaker identification, post-meeting reprocessing, and audio playback — all coordinated across three services.

The appetite is deliberately accepted before scoping begins. If the full scope doesn't fit, we cut scope — we don't extend the appetite.

---

Why Now

Live recording is the missing foundation for real-time AI value. File upload works but it delays every insight. The ML infrastructure (AssemblyAI, speaker diarisation, streaming insights) is already in place — this bet wires it to a live capture path. Without live recording, WordLoop remains a post-hoc tool. With it, it becomes a meeting partner.

---

Output

Check Bet Sizing to confirm the appetite judgment is realistic. Then move to The Pitch.