VerifyAI

A classifier that flags whether a phone call is a human or an AI-generated voice, built and shipped in one day.

At the Ivy College of Business hackathon I led a four-person team. VerifyAI targets deepfake phone scams, where a cloned voice of someone the target loves asks for money under pressure. We placed first in the technical track out of 28 teams.

Problem

Voice cloning is cheap, and the call arrives when the person will not inspect it calmly. Detection had to be simple enough to ship in a day and specific enough to be about the call, not a generic chatbot demo.

Process

We scoped a single path: ingest audio, transcribe it, and classify it using acoustic features and linguistic patterns as signals for synthetic speech. The team split capture, the model check, and the alert. Anything that did not help a person hang up or verify out of band was cut.

Solution

• Full pipeline from audio ingestion through speech recognition and transcription to classification.
• A plugin that scores a live call and surfaces an alert when the voice looks synthetic.

Results

• 1st Technical track
• 28 Teams in the field
• 4 People
• 1 day From scope to working product

Justin Moran · Software Engineer · Chicago, IL · LinkedIn · GitHub · Resume (PDF)