TruthShield AI

• 

  github code

• 

  run code

• 

  cloud server code

• 

  bot reply

Inspiration

Every day, thousands of innocent users fall victim to online scams, fake lottery notifications (like Jeeto Pakistan), and deceptive banking phishing links via WhatsApp and Telegram. Non-technical users cannot identify malicious links or manipulated transaction screenshots until it is too late. This critical gap inspired us to build TruthShield AI—bringing enterprise-level cyber security intelligence directly into everyday messaging apps for instant protection.

What it does

TruthShield AI is an automated, real-time cyber defense assistant operating inside WhatsApp and Telegram:Link & Text Scans: Instantly analyzes forwarded messages and links for fraud patterns. Screenshot Auditing: Uses advanced vision processing to scan uploaded transactional receipts or fraud banners for manipulation. Structured English Reports: Generates clean safety reports (Verdict, Threat Type, Risk Assessment, Safety Action) accompanied by automated text-to-speech voice notes.

How we built it

Backend Framework: Flask micro-framework hosted on PythonAnywhere middleware servers, with Streamlit driving the main dashboard. Artificial Intelligence: Integrated Google Gemini 1.5 Flash API for multi-turn semantic filtering and multimodal vision processing. Gateways: Leveraged Twilio WhatsApp Business API and Telegram Bot API as live response webhooks. Audio Synthesis: Integrated gTTS (Google Text-to-Speech) to dynamically compile threat metrics into crisp English voice notes.

Challenges we ran into

Environment Mismatches: Encountered server-side library path blockages (ModuleNotFoundError). We resolved this by injecting custom dynamic runtime paths (sys.path) to force-load local modules. Webhook Failures: Image data handling caused occasional backend proxy crashes (HTTP 500 error). We solved this by hardening data parsing layers and decoupling the core script architecture. Output Hardening: Forcing the AI model to output strictly formatted English data without conversational filler required extensive system prompt engineering.

AccDeploying a completely zero-dependency production build that achieves 100% stable uptime.

Creating a seamless multimodal pipeline that switches between text and image scanning within seconds over mobile networks. Transforming highly complex cyber intelligence tools into a simple, zero-installation chat interface.omplishments that we're proud of

What we learned

Hands-on expertise in configuring live webhooks, network routing architectures, and server-side traffic logging. The critical implementation of Lazy Loading inside function scopes to prevent server environment crashes. Engineering robust system prompts to control AI responses under volatile inputs.

What's next for TruthShield AI

Live Threat Database: Integrating international blacklists like VirusTotal and PhishTank to instantly block known malicious URLs. Localized Audio: Expanding the voice synthesizer into regional dialects to assist users who cannot read English text reports. Zero-Day Classification: Fine-tuning localized classifiers to detect complex, generative AI-crafted financial frauds before they spread widely.

Built With

• anti-fraud
• cloud-deployment
• computer-vision
• cyber-security
• flask
• generative-ai
• git
• google-gemini-api
• gtts
• natural-language-processing
• ocr
• python
• pythonanywhere
• streamlit
• telegram-bot-api
• text-to-speech
• twilio
• webhooks