A Digital Hand to Hold for the Vulnerable
Seguri is a privacy-first mobile security application designed to protect the most vulnerable users—the elderly, the non-tech-savvy, and the "digitally invisible"—from digital fraud.
While the tech industry races to build tools for the savvy and connected, the most vulnerable are often left defenseless. Seguri bridges this gap. It is a Hybrid-AI Security Suite that combines an Offline-First Neural Network for instant detection with Accessibility-First Design.
Why it matters: Security is a human right, not a subscription privilege. By removing the barrier of internet connectivity and technical literacy, Seguri restores agency to the vulnerable in developing regions where data is costly and connectivity is scarce.
Seguri provides a comprehensive defense system that works 24/7, mostly offline.
- What it is: A background service that never sleeps unless deactivated.
- How it works: Uses TensorFlow Lite to intercept incoming messages and notifications from SMS, Messenger, WhatsApp, Viber, and others.
- The Edge: Runs entirely offline. It blocks threats in milliseconds without needing Wi-Fi or Data, and preserves battery life by sleeping when inactive.
- What it is: A privacy-centric tool for checking suspicious emails or website text.
- How it works: Users paste text manually to be analyzed by the local AI.
- Privacy Guarantee: The analysis happens strictly on the device's RAM (Volatile Memory). The text is analyzed and instantly wiped; it is never uploaded to any server.
- What it is: A "Digital Paramedic" for when a user panics.
- How it works: An interactive, step-by-step emergency guide available 100% offline.
- Accessibility: Uses large buttons and reassuring color changes (Orange to Green) to guide the elderly through critical steps like "Turn off Internet" or "Check for Unknown Downloads" immediately after a hack.
- What it is: A live dashboard for awareness.
- How it works: When online, it uses Google Gemini to fetch real-time updates on the latest scam tactics prevalent in your specific country (e.g., "Sim Registration Scams" or "Holiday Phishing").
- Benefit: Keeps the user informed about new threats before they happen.
- What it is: An intelligent assistant that adapts to your connectivity.
- Offline Mode: Instantly answers common safety questions using Offline Ready Questions (Pre-set logic for fast, hardcoded safety advice).
- Online Mode: Connects to an LLM (Gemini AI) for specific, complex inquiries (e.g., "Is this specific URL format safe?").
- What it is: Context-aware protection engineered for the Global South.
- How it works: The scanning engine automatically loads TFLite weights tailored to the user's region.
- The Edge: Detects cultural nuances that western models miss.
- Example: Detecting "Taglish" (Tagalog-English) slang scams in the Philippines vs. Tax Fraud syntax in the USA.
- What it is: A digital paper trail.
- How it works: Automatically saves a history of detected threats and blocked messages using secure local key-value storage.
- Benefit: Provides read-only Local Storage evidence that users can show to banks or police authorities if fraud occurs, without the data ever leaving the phone.
| Feature | Technology | Connectivity | Privacy Status |
|---|---|---|---|
| Active Guard | TFLite + Local Heuristics | 📴 Offline | 100% Private (On-Device) |
| Manual Scanner | TFLite + Local Heuristics | 📴 Offline | 100% Private (RAM Only) |
| First Aid Kit | Interactive UI | 📴 Offline | 100% Private (Embedded) |
| Evidence Log | Shared Preferences | 📴 Offline | 100% Private (Local Storage) |
| Chatbot | Hybrid Logic | 🔄 Hybrid | Common Questions + Gemini AI |
| News Feed | Gemini AI | 📶 Online | Public Data Fetch |
To solve a human problem, we adopted a Privacy-First, Offline-First architecture.
- The Brain (Context-Aware Scanning Engine):
We trained multiple model weights using Python and TensorFlow.
- Global Layer: Trained on standard datasets (UCIML) for universal threats.
- Local Layer: Trained on localized datasets to detect cultural nuances (e.g., local slang).
- The Logic (Hybrid Detection): Built with Flutter, we defined our detection logic as a gated function that prioritizes human heuristics over raw probability to reduce false positives.
- The Cloud (Gemini): We engineered Gemini integration as a strictly optional layer. The core protection remains 100% offline.
Early in development, our AI was paranoid. It flagged innocent messages like "Grandma, I'm coming home, please cook dinner" as scams because it detected urgency. We learned that training on "Spam Only" datasets destroyed nuance.
The Solution: We rebuilt our dataset to include "Hard Hams"—safe messages that look like system alerts (e.g., OTPs, battery warnings)—and wrote complex Dart Heuristics to create a "Safety Net." We ensured that connection is never mistaken for corruption.
- True Offline Capability: Successfully optimized a Neural Network to run on budget phones with zero internet latency.
- 100% Privacy: Engineered a "RAM Only" Scanner where text is analyzed in volatile memory and instantly wiped.
- Inclusive Design: Replaced technical jargon with empathy. Instead of "Malicious Payload Detected," Seguri says, "Don't worry, you are safe."
Most AI security models are trained on Western datasets and English syntax. They fail to detect the nuanced, localized threats affecting the Global South.
Seguri bridges this gap. We are not just protecting devices; we are protecting dignity. By designing for the marginalized—those with poor eyesight, low literacy, or limited data plans—we ensure that the digital revolution lifts everyone up, rather than leaving our elders behind.
- Expanded Regional Support: Scaling the customization engine to support languages beyond English and Tagalog.
- Voice-First Interaction: Prototyping a feature powered by Gemini allowing the blind and illiterate to simply ask, "Is this safe?"
- Framework: Flutter (Dart)
- AI Engine (Local): TensorFlow Lite
- AI Engine (Cloud): Gemini AI API
- Local Storage: Shared Preferences
A visual tour of Seguri's interface and features
| 1. Getting Started | 2. Account Setup | 3. Regional Context |
|---|---|---|
 |
 |
 |
| 4. Home Dashboard | 5. Manual Scanner | 6. Threat Explanation |
|---|---|---|
 |
 |
 |
| 7. Scam First Aid | 8. Safety Affirmation | 9. Utility Tools |
|---|---|---|
 |
 |
 |
| 10. AI Chatbot | 11. Quick Questions | 12. Cyber News |
|---|---|---|
 |
 |
 |
| 13. Evidence Log | 14. Settings | |
|---|---|---|
 |
 |