Scam-Mah

What it is

Scam-Mah is a hackathon project that deals with scam callers so you do not have to. It scores incoming numbers to work out which ones look like spam, then lets an AI answer the ones that do, replying in a voice that keeps the caller on the line. The fun part was pointing a language model and a voice API at the problem instead of just blocking the number.

How it works

1. 1

   A scikit-learn model ranks phone numbers by how spammy they look. StandardScaler normalises the call history and an isolation forest flags the numbers whose hangup counts and spam reports sit far outside normal.

2. 2

   When a flagged call comes in, the recorded audio is uploaded to Gemini, which listens to the pitch and writes a reply in character.

3. 3

   ElevenLabs turns that reply into speech and plays it back down the line, so the caller hears a real voice answering.

4. 4

   A Flask app handles the audio round trip and a FastAPI endpoint serves the spam rankings to the web frontend.