AI FraudShield

• 

  AI FraudShield Multi-Agent Swarm Architecture Diagram

💡 Inspiration

The inspiration for AI FraudShield comes from a real-world multi-crore digital fraud incident right in my home region of Himatnagar, Gujarat, where victims lost over ₹14 Crores to sophisticated cybercriminals using fake stock market and digital arrest scams. Seeing vulnerable individuals and senior citizens fall prey to these high-pressure social engineering tactics made me realize that passive blocking is no longer enough. We needed an active-defense system that could neutralize scammers in real-time.

⚙️ What it does

AI FraudShield is an advanced AI-powered multi-agent framework designed to detect spam, phishing, and financial threats. It employs a two-tier agent swarm:

1. Sniffer Agent: Parses incoming suspicious texts, emails, or transcripts to instantly identify potential scams and red flags.
2. Baiter Agent (Dadaji Persona): If a scam is confirmed, this agent spins up an automated conversational honeypot mimicking a tech-illiterate Indian grandfather. It chats with the scammer using funny, confusing broken Hinglish and fake OTPs, successfully draining their operational time and keeping them away from actual victims.

🛠️ How we built it

The project is completely built using Python and Streamlit for a responsive, modern live dashboard interface. The intelligent multi-agent network is engineered using the Strands Agents SDK and connected to high-performance inference models via the Groq Cloud Infrastructure to guarantee ultra-fast response times.

🧠 Challenges we faced

One of the key technical hurdles was handling Python's updated object structure inside the modern Groq SDK when processing chat completions. We also spent significant effort fine-tuning the system instructions to ensure the Baiter Agent accurately maintains the 'Dadaji' persona consistently without dropping out of character, while ensuring the entire response fits under tight token budgets.

🏅 What we learned

Through this hackathon, we mastered orchestrating autonomous agent networks using the Strands SDK and learned how to leverage low-latency AI pipelines using Groq. Most importantly, we learned how to model AI personalities designed specifically for defensive honeypot scenarios.

Built With

• generative-ai
• groq-cloud-api
• multi-agent
• python
• strands-agents-sdk
• streamlit