Building Playcallr, a real-time sales coaching app

1. Build the mockupOn July 6, 2026 we built an interactive product mockup under the client's statement of work.
2. Build a clickable demoOn July 16, 2026 we built an end-to-end mobile demo with login, a lead queue, a branching live call simulator, post-call coaching, a playbook and practice drills.
3. Build the working appWe built the app with an invite-code join flow, owner and rep roles, and storage for calls, transcripts and prompt events. An offline outbox retries failed uploads.
4. Capture audio at the sourceWe rebuilt audio capture to take tab or system audio directly, plus a mic mode with echo cancellation off. A capability check tells reps what will work before the call.
5. Separate the speakersWe split meeting audio from the rep's mic to measure talk ratio, interruptions and playbook adherence, and to see whether the rep used the suggested line.
6. Detect objections fastKeyword-group matching on a rolling 14-second buffer runs in the browser in under 100 ms. Rules limit prompts to one at a time and 3 per minute, with an 18-second cooldown per objection.
7. Float above the meetingPrompt cards show in a floating window above Zoom using Document Picture-in-Picture.
8. Build AI coaching and phone callingWe built optional AI layers for paraphrased objections and nightly coaching notes, an AI cost ledger, and a Twilio phone call path. These are in the code and turn on per deployment; we have not confirmed they are enabled in production.
9. Test and tuneWe added a test call replay through the real detection engine, a trigger tuning bench, an eval harness and 24 test files.

Services in this project