Your Prince Is Not Real: Catching Scams While They're Still Happening

Description and use

YPINR detects scam calls and DMs in real time using a cheap first‑pass AI‑voice classifier and a deeper Grok classification, storing scam patterns in Postgres without personal data.

Problem

Scammers use cloned voices, phone calls, and DMs to trick victims before they can be detected by existing tools, which only flag bad URLs after a click.

Outcome

YPINR achieved real‑time scam detection across calls and DMs, reducing post‑incident losses and building a living encyclopedia of scam methods. The system was deployed live during a hackathon, earning a Top 10 finalist spot and the Best ElevenLabs Award.

Workflow

• 1. Capture first 30 seconds of a phone call with Outpost.
• 2. Send clip to ElevenLabs for AI‑voice scoring.
• 3. If score exceeds threshold, pass clip to Grok Voice Transcribe for full transcription.
• 4. Classify transcript with Grok 4 to determine scam type and confidence.
• 5. Store confirmed scam pattern in Postgres catalog (no personal data).
• 6. Optionally send plain‑text SMS warning via Textbelt.
• 7. In DMs, Sherpa Chrome extension highlights risky sentences on‑device.
• 8. Users can report new scam methods via the public hub.
• 9. Hub displays adaptive questionnaire and searchable scam catalog.

Evidence

YPINR is built around that gap, split into three surfaces sharing one "brain"

Source

We used ElevenLabs' speech classification as the first-pass screen on every call

Source

Our repo has a dedicated cursoragent contributor account that shipped real commits alongside the five of us

Source