Skip to content

Rakshak

Rakshak is a proactive AI intervention system that transforms AI from a passive assistant into an active digital guardian.

Code Video Cerebral Valley

Core trick

Watches the screen with Gemini multimodal input and speaks a Hindi warning only when a payment screen matches scam patterns.

Tags

securityfinancevoice

Description

Rakshak is a proactive AI intervention system that transforms AI from a passive assistant into an active digital guardian. Instead of waiting for users to ask if something is safe, Rakshak continuously observes high-risk digital interactions, understands the context using Gemini's multimodal reasoning, and intervenes only when it detects a potentially harmful decision. It explains why the interaction is risky, recommends the safest next action, and answers contextual follow-up questions to help users make informed decisions.
abhinandanmishra1/rakshak★ 10 forksTypeScriptCreated Jul 11, 2026 · last push Jul 11, 2026

README excerpt

# 🛡️ Rakshak — Proactive On-Screen UPI Scam Guardian

> *A second pair of eyes that only speaks to save you money.*

Rakshak is a real-time, proactive on-screen guardian built for the **Google DeepMind Bangalore Hackathon (Problem Statement 1: Real-Time Multimodal Interaction)**. It continuously watches a user's simulated phone payment screen, staying completely silent during normal operations, and proactively interrupts visually and verbally in the user's native language (Hindi) the instant it detects a deceptive scam.

---

## 🚀 The Problem & Vision

UPI transactions in India are growing at an exponential rate, but so is UPI fraud. Scammers routinely target elderly, rural, and first-time digital payment users using high-pressure scripts such as **"PIN-to-receive" scams**, **fake customer support verifications**, and **fraudulent cashback claims**. 

The fundamental issue is that these victims do not understand that entering a UPI PIN always **debits** money; it never receives it. 

### Why LLMs (Gemini) beat Hardcoded Rules:
1. **Generalization**: A new scam script or novel wording bypassing regex/keyword filters still gets caught by Gemini's zero-shot visual reasoning.
2. **Intent-vs-Action Mismatch**: Rakshak doesn't alert on normal merchant payments or utility bills. It reasons over the screen context to identify if the user was promised incoming credits but is performing an outgoing debit.
3. **Localized Explanation**: Rakshak speaks directly to the user in friendly, ur

Team