VeriCall

A “Mathematical Gatekeeper” to Protect Your Family from AI Scams.

• Voice is No Longer Proof

Phone calls are still the most universal way families communicate. But trust in voice is collapsing.

Today, scammers can clone a loved one’s voice from only a short sample.They exploit psychological panic — fake accidents, arrests, emergencies.

When people are scared, they cannot think clearly.We cannot protect families simply by saying “Be careful.”

The problem is not the phone itself — the problem is the lack of verifiable trust.

• The Solution: A Wall of Logic Without Emotions

  • VeriCall places a Mathematical Gatekeeper in front of every call.

  • Call → AI → Decision → TLSNotary → ZK Proof → On-Chain

    • Step 1: AI Screens Every Call First A real phone call comes in via Twilio. Before it reaches your family, the AI answers first. The AI has no emotions, no panic, no exceptions. Even if a scammer cries, screams, or fabricates an emergency, the AI stays calm and checks identity and facts using fixed screening rules embedded in public source code.

    • Step 2: Fixed Rules, No Voice Trust The AI collects information and follows strict rules. It never skips verification — even if the voice sounds exactly like your family. The decision is always: BLOCK (spam / scam) Or RECORD (important call)

    • Step 3: BLOCK or RECORD

      • If BLOCK → The AI stops the call immediately and sends you a spam alert.
      • If RECORD → The call is connected or you receive an email notification right away.

    • Step 4: Proof of Honesty with Web Proof + ZK

      • VeriCall does not ask you to “trust the AI.” Instead, every decision is mathematically sealed:

        • TLSNotary (Web Proof) cryptographically attests the Decision API response
        • RISC Zero ZK Proof compresses it into a succinct proof

      • The result is stored in a smart contract on-chain. Each record locks together:

        • Input (transcript hash)
        • Output (BLOCK / RECORD + reason)
        • Logic (systemPromptHash + sourceCodeCommit)

      • No one can rewrite the story afterward.

• Verifiable Trust → Explainable AI

  • VeriCall is not a black box. Every on-chain record provides an audit trail: - What decision was made - Why it was made - Which rules were applied - Which exact GitHub commit contained the screening logic
  • You can verify everything later — independently.

• Our Message - “Protecting family love with a shield of technology.“ --- This system can stop: “Grandparent Scams“ in the US, “Impersonation Scams“ in Europe, “It's me Scams(OreOre)“ in Japan

  • In a world where we cannot trust voices, VeriCall creates a new standard: Verifiable Trust
  • “Your call data, your rules.“ - In Web2, telecom companies own your call records. - In VeriCall, you own the proof. You decide whether to share it with police, insurance, or no one. This is data sovereignty for the AI age.

VeriCall is built as an end-to-end verifiable AI call screening pipeline.

The frontend and backend are written in TypeScript using Next.js, deployed on Google Cloud Run, with real phone calls handled through Twilio Media Streams. Incoming calls are answered first by an AI receptionist powered by Vertex AI Gemini 2.5 Flash, combined with Google Speech-to-Text and Text-to-Speech.

The core contribution is not just AI classification, but making every decision mathematically non-repudiable:

The AI produces a deterministic decision (BLOCK / RECORD) with an explanation.

A Decision API returns structured JSON containing the decision, reasoning, transcript hash, system prompt hash, and the exact Git commit SHA of the screening rules.

TLSNotary (vlayer Web Prover) cryptographically attests that this JSON truly came from VeriCall’s server, without revealing private call contents.

RISC Zero ZK proofs (vlayer ZK Prover) compress the web proof into a succinct proof.

A Solidity smart contract (VeriCallRegistry) verifies 15+ integrity checks, including decision–journal binding (you cannot submit a proof for “RECORD” and store “BLOCK”).

All decisions are anchored on-chain with Input + Output + Logic sealed together, creating a public audit trail for explainable and accountable AI call protection.