HamCaller

AI-powered spam call detection model. Instantly classify phone call transcripts as SPAM or LEGITIMATE.

Quick Start

# Pull the model
ollama pull hamcaller

# Test it
ollama run hamcaller "Your car warranty is expiring. Press 1 to renew."
# Output: SPAM

What it Does

HamCaller detects spam calls including: - Extended warranty scams - Fake IRS/government calls - Prize/lottery scams - Robocalls and phishing attempts - Tech support scams

It recognizes legitimate calls like: - Medical appointments - Delivery notifications - Personal/family calls - Known business contacts

Usage

# Simple classification
ollama run hamcaller "Hi mom, checking in to see how you're doing."
# Output: LEGITIMATE

# In your app
ollama run hamcaller "This is the IRS. You owe taxes immediately."
# Output: SPAM

Model Details

Base: Gemma 3 (1B parameters)
Size: 815 MB
Training: 185 real phone call examples (165 spam, 20 legitimate)
Output: Single word - “SPAM” or “LEGITIMATE”
Speed: ~2-3 seconds per classification

Example Integration

import subprocess

def is_spam(transcript):
    result = subprocess.run(
        ["ollama", "run", "hamcaller", transcript],
        capture_output=True, text=True, timeout=30
    )
    return "SPAM" in result.stdout.upper()

# Use it
is_spam("You've won a free cruise!")  # True

Privacy

100% local processing - no data leaves your machine.

Built with Gemma 3 • Trained on real spam call data • Optimized for accuracy

Models

Readme