Android · On-device AI · by CyberTech Innovators

Stay a step ahead.

ViraShield checks suspicious links, QR codes and messages before you act — and tells you exactly why something looks dangerous, in plain language.

0 bytesmessage content uploaded
~39 KBon-device AI model
6Link DNA strands
ViraShield home dashboard
Phishing blockedpaypa1-verify.xyz
Link is safeOpening in Chrome
Scam SMS flaggedAsks for OTP
01

Problem statement

The risk behind an urgent message.

Scam messages pressure people to judge unfamiliar links and requests in seconds, with almost no context. One wrong tap can cost accounts, privacy and money.

₹0 cr Cyber-fraud amount reported in India during 2025
0 lakh Financial-fraud complaints on NCRP in 2025
0% Complaint growth from 2024 to 2025

Source: MHA / PIB, 11 Feb 2026. Reported national figures across cyber fraud — not confirmed losses or Android-specific cases. Growth calculated from 19,18,835 (2024) → 24,02,579 (2025) complaints.

02

Proposed solution

Protection built into the Android workflow.

ViraShield focuses on the one moment that matters — before you click a link or share a code.

01

Choose your sources

Approve ViraShield as the default browser in Android's own role dialog, then opt in to checking new SMS or notifications from apps you pick. Message sources start off.

02

Links & messages get checked

Every link routed through ViraShield is scored before it opens. Message text is analysed on the phone by a local AI model plus safety rules.

03

Read the reasons, then decide

Safe links open in your preferred browser. Risky ones are blocked with plain-language reasons. The final call is always yours.

Tap a link
ViraShield intercepts
Blocklist · Heuristics · Link DNA
Safe → opens in your browser
Risky → blocked with reasons
✦

Features

Everything inside ViraShield.

24 tools in one app — a complete security kit for everyday Android users.

Automatic link interception

Acts as your default browser and checks every HTTP/HTTPS link you tap, system-wide — then forwards safe links to the browser you actually use. Chrome, Firefox, anything.

Core

Link DNA analysis

A risk profile built from six independent strands, each with a plain-language reason:

  • Domain age
  • Redirects
  • URL structure
  • Certificate
  • Look-alike similarity
  • Threat history

Heuristic risk score

0–100 on-device scoring: IP hosts, punycode/homographs, excessive subdomains, suspicious TLDs, hyphen-heavy domains.

Fake-site detector

Compares what a page claims to be — title, logo, favicon, login form — with the domain it's really served from.

Local threat blocklist

Background-synced domain database with SHA-256 hashed lookups via WorkManager.

Block screen

Dangerous sites stopped with a clear warning, strong haptic alert and a "continue at your own risk" check.

Protection modes

Strict, Balanced or Lenient — plus custom rules and trusted domains.

Link preview & scan report

A safety badge and page details before opening, and a full breakdown of every check after.

Manual link analysis

Paste or type any URL and inspect it before you open it.

QR scanner

CameraX + ML Kit reveal the real destination behind a QR code and check it first.

Clipboard monitor

Optional background service that auto-scans URLs the moment you copy them.

Safe Browser

Built-in browsing view with an ad and tracker domain blocklist.

Incoming SMS checks

With your permission, new SMS are checked on the phone for scam signs. Your existing inbox is never read.

App notification checks

Checks visible message notifications from apps you choose — WhatsApp, Gmail and more. Duplicates suppressed for 10 minutes.

Check a message

Paste or type any message you received for an instant on-phone check. Nothing is sent anywhere.

Smart safety rules

Catches OTP/secret-code requests, payment pressure, urgency, remote-access apps, impersonation and risky links.

Charging check

Detects a USB data port vs. a pure wall charger and warns you about juice jacking on untrusted PCs.

App Watchdog

Flags risky permission combinations, unusual install sources and changes since the last scan — metadata only.

Threat DNA Network

On-device graph of domains, IPs, redirect hops and imitated brands — revealing connected threat clusters.

Threat radar

Explore a visual threat-intelligence map straight from the dashboard.

Simple interface

One switch: a large-text home with just three actions — Check a link, Message protection, Settings.

Phishing quiz & tips

"Is this URL safe or suspicious?" — learn to spot scams yourself.

Dashboard & reports

Protection score gauge, weekly chart, daily summaries and weekly report notifications.

Themes, haptics & widget

Five colour themes, distinct vibration patterns, a home-screen widget and a floating scanner bubble.

▶

Interactive demo

Scan a link. Right here.

A taste of ViraShield's on-device heuristics, running in your browser. Nothing you type leaves this page.

Try:

Website demo of a few URL heuristics only. The ViraShield app also uses its local blocklist, Link DNA, certificate and redirect checks and the fake-site detector. No tool can guarantee a link is safe.

Privacy by design

Your messages never leave your phone.

Local message analysis

All message inference runs on the phone. No message content is uploaded.

Nothing original saved

Message bodies and sender details are processed in memory only.

Minimal history

Max 20 warning summaries, kept 7 days, excluded from backup. Clear anytime.

Permission-based

You pick each source. Revoking access stops checks instantly — even queued ones.

Hands off

Never sends, deletes or marks messages read. Never opens extracted links.

No bypassing

Hidden, muted or redacted notification content is respected, not worked around.

Message protection screen
On-device · No cloud
03

Tech stack, AI & tools

A lightweight local AI engine.

virashield/local_model
~39KB

Bundled English spam classifier

  • algorithmMultinomial Naive Bayes
  • learned_weights2,400
  • exact_size39,484 bytes
  • runs_on100% on-device
  • window16k chars · 4k overlap
  • extra_rulesOTP · payment · urgency · remote-access · impersonation · URL

A text classifier, not a generative chatbot. Trained on the UCI SMS Spam Collection (Almeida & Gómez Hidalgo, CC BY 4.0).

Android app

Kotlin · AndroidX · XML + ViewBinding · Material · Android 11+

Link checks

Heuristic scanner · Link DNA · local blocklist · RDAP · OkHttp · SHA-256

QR & camera

CameraX · Google ML Kit barcode scanning

Background work

WorkManager · coroutines · NotificationListenerService · SMS receiver

Platform APIs

RoleManager · foreground services · app widget · haptics

Model training

Python training & evaluation pipeline

04

Prototype demo

The working prototype.

Real screenshots from the ViraShield Android app. Drag to explore.

0%Spam precision
0%Spam recall
0Held-out messages evaluated · 31 abstentions
0Automated tests passing (17 JVM + 14 Android)
106True positive
3False positive
901True negative
18False negative

Historical English SMS dataset results — not real-world scam accuracy. Real-device Gmail/WhatsApp delivery and modern scam patterns still need validation. Debug build and lint pass with zero errors.

05

Audience & impact

Everyday users, meaningful reach.

0B Internet subscriptions in India, March 2026 (TRAI)

Everyday Android users

People who need clear guidance before trusting a link or message.

Families helping relatives

Simple interface and large text for seniors and first-time smartphone users.

Campus pilot communities

Students and schools where we can run opt-in pilots.

Intended impact

Help people pause and understand warnings. We'll measure comprehension, false alerts, setup completion and repeat use — before claiming any financial outcomes.

06

Future plans & scalability

Where we go next.

Go-to-market

01

Pilot

Recruit opt-in campus and community groups.

02

Refine

Use consented feedback to improve onboarding and warnings.

03

Distribute

Test consumer channels and institutional sponsorship.

Business hypothesis: free core checks, with paid convenience or sponsored access.

Roadmap

Near term

Real-device pilots

Evaluation against current English scam patterns.

Next

Regional languages

Testing in Indian languages and more accessible onboarding.

Later

Wider distribution

Vetted threat-intelligence updates at scale.

On-device inference keeps per-message cloud cost near zero — ViraShield scales with users, not servers.

We're seeking

Pilot access & cybersecurity mentorship.

Meet the team

The team

CyberTech Innovators.

Two student builders on a mission to help people make a safer decision before they act online.

MP

Mukhil Pranesh K

Co-founder · Developer

CA

Chanjiv A

Co-founder · Developer

Mukhil Pranesh K and Chanjiv A of CyberTech Innovators
?

FAQ

Good questions.

Why does ViraShield need to be my default browser?

That's how Android lets an app see links before they open. ViraShield checks the link, then hands safe ones to the browser you prefer — you keep browsing with Chrome, Firefox or anything else. You approve it in Android's own dialog.

Does ViraShield read all my messages?

No. It only checks new SMS (after you allow it) and visible notification text from apps you select. It can't read full chat or email histories, attachments or hidden content, and it never stores original messages.

Is my data uploaded to the cloud?

Message analysis happens entirely on your phone. Some link features (like domain registry lookups or blocklist sync) use the network, but message content is never uploaded.

Can ViraShield guarantee a link or message is safe?

No security tool can. ViraShield flags suspicious patterns and explains why — it can miss scams or flag genuine messages, and it can't verify who a sender really is. When unsure, call someone you trust.

Which Android versions and languages are supported?

Android 11 and above. The message model currently supports English; regional-language testing is on the roadmap.