Voice Notes to Personal CRM - Builds & Demos | Soto Systems Lab

➔ Build Specifications

👥

Who It Helps

Operators, small business owners, and mobile sales professionals.

⚠️

Problem

Reps have great ideas or meeting take-aways on the road, but they delay entering them because typing into a CRM app on a phone is slow and annoying.

⚙️

How It Works

User records a voice note in Telegram. A webhook sends it to n8n, which transcribes it via Whisper, extracts fields using Claude, and inserts/updates the contact in a Notion database.

🧠

What I Learned

User friction must be absolute zero. Pushing a single button in Telegram is the only reason this system gets 100% adoption.

🚀

What I'd Improve Next

Add voice-matching checks so only verified user voices can trigger updates to the database.

🛠️

Tools Used

Telegram Bot APIn8nOpenAI WhisperClaude 3.5 SonnetNotion API

This project solves the "lazy logger" problem by removing the need for an app UI. The interface is a chat app you already have open.

Problem

Reps have great ideas or meeting takeaways on the road, but they delay entering them because typing into a CRM app on a mobile phone is slow and annoying.

Solution

An automated pipeline that lets a user speak their mind into a Telegram message and have it parsed and categorized automatically.

How It Works:

Record: User records a voice note in the Soto Systems Telegram Bot.
Translate: Webhook triggers n8n, which sends the audio file to Whisper for transcription.
Parse: Claude reads the transcript and identifies fields like Contact Name, Company, Follow-up date, and Meeting Notes.
Log: Writes directly to the Notion CRM database.

What I Learned

Zero Interface is Best: A chat interface is the most adopted interface. Do not build a custom mobile app when you can use Telegram or WhatsApp.
Semantic Mapping: The LLM is highly resilient at understanding messy transcriptions (e.g. mapping "John at globex" to "John Doe at Globex Corp").