Inspiration

Digital scams are becoming increasingly common. People receive fraudulent SMS messages pretending to come from banks, government departments, delivery services, and job recruiters. A single misleading message or suspicious link can result in financial loss or stolen personal information.

We wanted to build something that could protect people before they become victims. Instead of simply warning users about suspicious messages, we envisioned a mobile-first security application that actively identifies potential threats and separates them from normal conversations.

That idea inspired ScamShield AI — an intelligent, privacy-conscious SMS protection system designed to make digital communication safer.

What It Does

ScamShield AI analyzes incoming SMS messages to identify potential scams, spam, phishing attempts, and suspicious links.

Its core features include:

  • AI-powered message classification: Analyzes message content to identify suspicious patterns.
  • Smart Inbox and Threat Vault: Keeps normal messages in the Inbox and moves detected suspicious messages into a separate Vault.
  • Offline detection: Uses an on-device machine learning model and rule-based analysis to work without depending on a constant internet connection.
  • URL analysis: Extracts links and evaluates suspicious URL characteristics.
  • Multilingual awareness: Includes detection support for English, Tamil, and Tanglish patterns.
  • Risk assessment: Assigns risk levels to messages and provides explanations and safety recommendations.

How We Built It

We developed ScamShield AI as a native Android application using Kotlin and Jetpack Compose.

The application combines machine learning with a rule-based threat detection engine. Its detection pipeline follows these steps:

  1. An incoming SMS is received by the application when it is configured as the device's default SMS app.
  2. The message is processed through language detection and text preprocessing.
  3. URLs are extracted and analyzed for suspicious characteristics.
  4. A lightweight TF-IDF and logistic regression model performs local message classification.
  5. A rule-based risk engine evaluates additional indicators, including urgency, OTP requests, impersonation, payment pressure, and suspicious job offers.
  6. The final risk assessment determines whether the message belongs in the normal Inbox or the Threat Vault.

We also use Room Database to manage message records locally and maintain the application's data.

What We Learned

Building ScamShield AI helped us understand how real-world mobile security systems combine multiple approaches rather than depending on a single machine learning prediction.

We learned about:

  • Android SMS handling and default SMS application requirements.
  • On-device machine learning and offline inference.
  • Text preprocessing, TF-IDF, and logistic regression.
  • URL feature extraction and suspicious-link analysis.
  • Rule-based threat detection and risk scoring.
  • Local data management using Room Database.
  • Designing a mobile security experience that separates threats from everyday communication.

Challenges We Faced

One of our biggest challenges was designing a detection system that could distinguish genuine messages from fraudulent ones. Not every message containing words such as OTP, bank, or urgent is a scam. We therefore combined machine learning predictions with contextual rules to improve the decision-making process.

Another challenge was implementing Android's SMS handling correctly, particularly configuring the application as the default SMS app and processing incoming messages reliably.

We also had to design the system to operate offline, which meant keeping the core classification and risk-assessment pipeline on the device instead of relying entirely on cloud APIs.

Our Vision

ScamShield AI is more than a spam filter. Our vision is to develop a practical, privacy-conscious digital safety companion that helps people recognize threats before they take action.

We aim to continue improving its detection accuracy, multilingual capabilities, suspicious-link intelligence, and protection against emerging scam techniques.

ScamShield AI — Detect the Threat. Protect Your Trust.

Built With

  • android
  • android-studio
  • artificalintelligence
  • cybersecurity
  • jetpackcompose
  • kothlin
  • logisticregression
  • machine-learning
  • mobilesecurity
  • naturallanguageprocessing
  • offlineai
  • phishingdetection
  • python
  • roomdatabase
  • rule-basedsystem
  • sms
  • spamdetection
  • sqlite
  • tf-idf
  • urlanalysis
Share this project:

Updates

Submission history