Clawckathon 2026 was a hackathon Resemble co-sponsored with Telnyx and Lovable, held March 20 to 22. HughKnew, a tool that teaches families to catch AI voice scams before they happen, took first place using Resemble Detect and Resemble Identity as a core part of their tool.
THE PROBLEM
Why nobody could practice for this before
Americans age 60 and older lost $7.7 billion to online fraud in 2025, according to the FBI's IC3 Annual Report. AI voice cloning is a growing piece of that. It now takes as little as three seconds of audio to clone a voice convincingly enough to fool someone who's never heard of a deepfake and has no reason to doubt the voice on the phone.
Every member of the team had a version of the same story: a parent, a grandparent, an older relative who was capable and sharp but unprepared for a technology that didn't exist the last time they needed to be skeptical of a phone call. The team, Dan Klosterman, Noah Vandal, and Isaiah Klosterman, decided from the start that nobody should have to download an app or learn a new interface. Every interaction happens over a real phone call or a text message, because that's the technology their users already trust.
THE SOLUTION
The four features they built
HughKnew is built around four pieces that work together:
Voice Circle. Trusted contacts (a grandson, a daughter, a caregiver) record a short voice sample once. That sample becomes a verified reference by Resemble Identity, so the next time "James" calls, there's something to check him against.
Scam Training. HughKnew places a practice scam call using a cloned version of a trusted contact's voice, at an adjustable difficulty level, and watches what the user does. It's the closest thing elder-fraud prevention has had to a rehearsal: you can't practice against a real scammer, so HughKnew builds one you can.
Deepfake Detection. If a call feels off, the user can upload the recording directly and get a real-or-fake read against the verified voice profile. Resemble's detection runs underneath a screen built for someone who has never heard the word "deepfake."
Companion Calls. A daily check-in call from "Hugh," an AI companion, tracks verbal fluency, memory recall, and conversation flow over time, watching for early signs of cognitive decline. If something looks off, it lets trusted contacts know. It wasn't part of the brief. The team built it anyway.
HOW IT'S BUILT
The stack behind it
Stitched together in a single weekend:
- Telnyx: Call Control API, AI Assistants, voice cloning, and the LLM layer powering the calls themselves
- Resemble AI: voice identity enrollment and deepfake detection for the Voice Circle and detection features
- ClawdTalk + OpenClaw: the companion assistant, with scheduled calls run through the Events API
- ClawHub: the team also built and published their own Persona Plugin for persistent caller memory across calls
- Supabase: auth, Postgres, storage, and realtime
- Lovable: the frontend, built high-contrast and low-clutter for older users, and responsive enough to work on whatever device a family member already owns
Every feature runs live against real APIs and that's one of the many reasons this project won.
THE RESULT
What the demo shows
The strongest argument for HughKnew is the demo itself. In the first practice call, the user recognized the pressure tactic ("I need you to send me money right away") and hung up. In the second, run at a harder difficulty, she started reading off where to send the money before she caught on.
That gap, between the call you catch and the one you almost don't, is what the product is built around: real risk, practiced safely, until the pattern is familiar enough to catch when it matters.
WHAT'S NEXT
A weekend build that already works
HughKnew is live and working today. The app and code are both public, and every feature in the demo runs the same way for anyone who signs up.
Live app: hughknew.lovable.app
Code: github.com/edgeteamdan/HughKnew
PRODUCTS USED
Resemble Detect / Resemble Identity / Audio Deepfake Detection
