Nov 21, 2026

Guardian

#agora #murfai #agenticai #frauddetection voiceagent

Guardian is a voice-first scam response and recovery assistant. It is designed for someone who has just received a threatening call, noticed a suspicious transaction, or shared information with a possible scammer. In that moment, the person may be frightened and unsure what to do first. Guardian’s purpose is to help them understand the situation, make a practical plan, and keep control of every consequential action.

The main demonstration uses a digital arrest scam: a caller pretends to be a police officer, threatens the user, and demands ₹35,000. The user speaks naturally with Guardian through Agora Conversational AI and hears its replies through Murf AI Falcon. Guardian asks questions to establish the facts, such as whether money was transferred, whether the transaction is recognized, and whether banking credentials were exposed. It should avoid making assumptions or asking for an OTP, PIN, password, or full card details.

Guardian then brings together context and action. It can retrieve the user’s preferences and incident history, inspect transactions from a separate synthetic bank, assess the situation, and present an inspectable response plan. Its registered tools can create an incident, request a demo card freeze, prepare a trusted-contact notification, schedule follow-ups, and generate a report. This is the project’s central goal: the conversation should lead to coordinated, recorded work rather than end with general advice.

The user remains in charge. Freezing the synthetic card or sending information to a trusted contact requires an explicit approval in the dashboard; saying “yes” to the voice agent cannot bypass that gate. After an approved action, Guardian checks the resulting state before reporting success. If a provider fails or the result cannot be verified, it should say so plainly. Incidents, action results, approvals, and follow-ups are saved so the user can return later and see what happened.

Google integration is account-specific: a user can connect their own Google account to support approved Gmail notifications and Calendar follow-ups, and choose their own trusted contact. These settings live in the profile area. The follow-up plan includes general check-ins and, when the user supplies confirmed bank complaint details, conditional reminders to review applicable RBI milestones. A reminder is guidance to check progress; it does not guarantee a refund or file a complaint.

The intended end-to-end journey is context → conversation → plan → approved action → verified outcome → follow-up. For the hackathon, success means demonstrating that journey with real voice interaction and at least two coordinated tool calls that complete and verify a meaningful task. The bank and card actions are explicitly a sandbox: Guardian cannot freeze a real card, reverse a transfer, or report a case to authorities. Its role in a real incident is to help the person act promptly through their bank’s and official reporting channels without overstating what the application itself has done.

This build was uploaded as a hackathon project

Hackathon

Build with Agora - BLR

View All Projects
Give a star to encourage!Discussion
Start a new conversation!
Login to join the discussion