Our app allows users to scan dubious content to determine whether it's AI generated and/or malicious.
Hestia provides the user an archive of previously flagged content, such as websites, photos, videos, and calls
Hestia gives families recommendations on what security actions to take so they can keep everyone protected in case an attack has been made.
Hestia's home page gives the user a rundown of recent actions taken.

Hestia — AI Scam & Threat Detection

Inspiration

Every day, millions of scam attempts target individuals across the world, many of them increasingly powered by AI.

A parent receives a call from someone impersonating their child
A mother receives a FaceTime call from a deepfake version of her daughter, urging her to open the door
Voice cloning and synthetic media are rapidly lowering the barrier for fraud

These are no longer hypothetical scenarios; they are happening now.
Hestia was built to address this emerging reality.

What It Does

Hestia acts as a background guardian on a user’s device, proactively identifying and flagging AI-driven threats before they cause harm.

Core Features:

Real-time Call Monitoring
Intercepts and analyzes calls, warning users when a conversation may involve a fraudulent or AI-generated actor
AI Threat Detection Engine
Uses multiple detection algorithms to identify patterns consistent with scams, voice cloning, and synthetic media
User-Friendly Dashboard
Designed for accessibility, especially for users with less experience in modern technology
Media and Link Scanning
Allows users to upload images, videos, or suspicious links from their gallery or browser for analysis

How We Built It

We brainstormed for hours a couple of days before the hackathon, building on the "Future" theme and ranking ideas by impact and by which problems we believe should not exist in the future. We landed on scam prevention, recognizing that AI voice fraud is becoming more realistic and more common. While older adults are especially vulnerable, this is a product that can protect anyone.

We arrived at the hackathon with that direction, met Kenji on-site, and formed a four-person team with clearly defined roles:

Mason — Backend and infrastructure
Alef — UI/UX design
Kenji — Detection algorithms
Dennis — Triage, testing, and business framing

From the commit and push history, development progressed in tight iterations across April 24–25, 2026:

Development Phases

Initial Setup
Project scaffolding and architecture documentation
Core Infrastructure
Bot caller implementation and Supabase Realtime signaling
Application Foundation
React Native app structure and iOS native bridge wiring

Parallel Feature Development

After establishing the foundation, we split into feature branches and merged continuously:

Detection Systems (Kenji)
- ElevenLabs Layer 1 detection
- Core ML Layer 3 classifier
Frontend (Alef)
- UI migration
- Screen design and polish
Backend and Reliability (Mason & Dennis)
- Call-flow hardening
- iOS build and signing fixes
- PushKit and VoIP reliability updates
- Calibration and text-risk pipeline development

Final Result

By the end, we achieved a working demo loop:

Incoming VoIP call
Real-time on-device analysis
Cumulative risk scoring
Local alert: “Caller may be AI” triggered once the risk threshold is exceeded

Challenges We Ran Into

Designing detection algorithms that could run efficiently on mobile hardware
Building a stable real-time audio processing pipeline
Managing development across Android Studio and Metro, which required frequent reloads
Integrating voice cloning tools such as ElevenLabs, which introduced latency and complexity