Inspiration
Online scams are becoming more sophisticated every day, especially through messaging apps, fake payment requests, phishing links, and social engineering tactics. In Indonesia, many victims are not necessarily “careless”, they are often manipulated under panic, urgency, or fear.
We wanted to build something more than just another AI chatbot.
That’s why we created Amanin, an autonomous AI scam investigation agent designed to help everyday people identify suspicious messages, links, and payment requests before it’s too late.
Our inspiration came from a simple idea:
What if everyone had an AI companion that could pause, investigate, reason, and explain potential scams in real time?
What it does
Amanin is an autonomous multi-agent AI system that investigates suspicious digital activity.
Users can upload:
- WhatsApp screenshots
- Emails
- SMS messages
- URLs
- Marketplace chats
- Payment-related screenshots
Amanin then autonomously:
- Extracts and analyzes content
- Detects social engineering patterns
- Investigates suspicious domains and links
- Evaluates payment-related risks
- Aggregates findings across multiple agents
- Generates a human-friendly risk explanation
Instead of simply saying “this is a scam,” Amanin explains:
- why it is suspicious
- what manipulation tactics are used
- how confident the system is
- what users should do next
The platform is designed to be understandable even for non-technical users and elderly users.
How we built it
We built Amanin using a multi-agent architecture powered by Hermes Agent.
Tech Stack
- Hermes Agent
- Vision/OCR models
- Domain reputation APIs
- WHOIS lookup tools
Agent System
Amanin consists of several specialized agents:
Orchestrator Agent Coordinates investigation workflows and delegates tasks dynamically.
Threat Analyzer Agent Detects psychological manipulation tactics such as urgency, fear, fake authority, and OTP requests.
Domain Investigator Agent Analyzes URLs, domain age, reputation, typosquatting, and phishing indicators.
Identity Validator Agent Evaluates suspicious phone numbers, emails, and payment-related entities.
Risk Assessor Agent Aggregates findings into a final explainable risk score.
Explainer Agent Converts technical findings into simple and understandable language.
We also designed dedicated system files such as:
AGENTS.mdSOUL.mdWORKFLOW.mdTOOLS.mdSAFETY.md
to define agent behavior, communication rules, investigation philosophy, and safety boundaries.
Challenges we ran into
One of the biggest challenges was balancing:
- autonomy
- explainability
- and speed
Scam detection is highly contextual. Some messages may look legitimate but contain subtle social engineering patterns. We had to ensure the agents could reason across multiple signals instead of relying on simple keyword matching.
Another challenge was creating a system that feels trustworthy and calm rather than fear-inducing. We wanted Amanin to guide users clearly without creating unnecessary panic.
Coordinating multiple agents within a limited hackathon timeframe was also challenging, especially ensuring consistent structured outputs between agents.
Accomplishments that we're proud of
We are proud that Amanin is not just a “chatbot wrapper,” but a truly autonomous investigation workflow.
Some accomplishments we are especially proud of:
- Building a functional multi-agent orchestration system within 12 hours
- Creating explainable scam analysis instead of black-box outputs
- Implementing dynamic tool usage and investigation routing
- Creating a strong safety-first AI personality and communication system
We are also proud of Amanin’s human-centered design philosophy:
protecting people without blaming them.
What we learned
This project taught us that AI agents become significantly more powerful when they:
- collaborate
- specialize
- and reason together
We also learned that explainability is critical for trust. Users are more likely to trust AI systems when they can understand why a decision was made.
Another important lesson was that AI safety and emotional UX matter just as much as technical capability, especially when dealing with scams, fear, and financial risk.
Finally, we learned how powerful autonomous workflows can become when combining:
- reasoning
- tool usage
- memory
- and structured orchestration
What's next for Amanin
We believe Amanin has strong real-world potential beyond the hackathon.
Our future plans include:
- Real-time WhatsApp and email integrations
- Browser extension support
- Community-driven scam pattern memory
- Multi-language support
- Voice-based scam analysis for elderly users
- Payment fraud detection integrations
- Live threat intelligence feeds
- Personalized safety profiles
We also want to explore:
- proactive scam prevention
- AI-powered family protection modes
- and nationwide digital literacy initiatives
Our long-term vision is simple:
Make digital safety accessible to everyone through autonomous AI agents.
Built With
- hermes
Log in or sign up for Devpost to join the conversation.