Project
Salescribe
Voice memos for sales calls, with an AI that actually listens.
Built with Next.js, TypeScript, Claude, OpenAI Whisper, Firebase, Tailwind.
What it is
Salescribe is a voice memo app built for the moments right after a sales call, when everything is fresh in your head and you have about ninety seconds before the next thing pulls you away. You hit record, talk for a minute or two, and Salescribe does the rest.
Whisper transcribes what you said. Claude reads the transcript and pulls out the structured deal information you’d normally have to type into a CRM by hand. Then it follows up with the coaching questions you probably should have asked yourself: what’s the actual next step, who else needs to be in the room, what’s the real objection. When you have a meeting on the calendar with that account, Salescribe stitches all your prior memos into a pre-meeting briefing so you walk in with the context already loaded.
Why it matters
Most CRM workflows lose at the friction step. Reps don’t update the system because typing on a phone after a call is the worst. Salescribe inverts that: talk first, structure later, with the AI doing the structuring.
The coaching piece is the part I’m most excited about. It turns each call into a small learning loop, prompting you to interrogate your own assumptions while the memory is fresh.
Tech notes
- Frontend: Next.js 16 (App Router) with TypeScript and Tailwind v4.
- Transcription: OpenAI Whisper.
- Extraction + coaching: Claude Sonnet 4.6.
- Text-to-speech: OpenAI TTS for the coaching follow-up questions.
- Auth + data: Firebase Authentication and Cloud Firestore.
- Hosting: Firebase App Hosting (running on Cloud Run).
- Bonus: uses the browser’s SpeechRecognition API for real-time feedback while you talk.