ScamShield AI — A Conversational Safety Assistant

• 

Inspiration

Online scams are becoming harder to recognize. A message can look like it came from a bank, a company, a delivery service, or even someone we know. Many people know that scams exist, but when they receive a convincing message, it can still be difficult to decide what is real and what is not.

I wanted to build something that doesn't simply tell people "this is a scam." Instead, I wanted an assistant that could have a conversation with the user, understand the situation, explain the warning signs in simple language, and help them decide what to do next.

That idea led to ScamShield AI — A Conversational Safety Assistant.

I chose the Alexa+ track because I wanted to explore how conversational AI and tool-based AI systems can be used to solve a real-world problem rather than building another simple chatbot.

What it does

ScamShield AI is designed to help users investigate suspicious messages and potential online scams through natural conversation.

A user can describe or provide a suspicious message and ask something like:

"I received a message saying my bank account will be blocked unless I click this link. Is it genuine?"

Instead of only returning a yes/no answer, ScamShield AI is designed to:

• Identify potential warning signs.
• Explain why those signs may be suspicious.
• Analyze available information using specialized tools.
• Help the user understand the level of risk.
• Recommend safer next steps.
• Be transparent when something cannot be verified.
• Support a more natural, conversational interaction.

The goal is not to replace cybersecurity experts or banks. The goal is to give people a helpful first layer of guidance when they are unsure about something they received online.

How we built it

We designed ScamShield AI around a conversational AI architecture with an MCP-based backend.

The core idea is to give the AI access to specialized tools instead of expecting the language model to do everything itself.

The planned architecture includes:

User → Conversational Interface → AI Assistant → MCP Server → Safety Tools → AI Response

The MCP server provides tools that can be used when additional analysis is required. These can include message analysis, suspicious-link analysis, and generating practical safety guidance.

We are using Python for the backend and exploring MCP to understand how AI assistants can discover and interact with external tools.

We are also designing the system with an important principle in mind: the assistant should not pretend to know something it cannot verify.

For example, if a message looks suspicious but there is not enough evidence to confirm that it is fraudulent, the assistant should explain the uncertainty rather than confidently calling it a scam.

Challenges we ran into

One of the biggest challenges was deciding how to make ScamShield more than a simple AI chatbot.

It is easy to ask an AI model whether a message looks suspicious. The harder problem is creating a system where the assistant can decide when it needs additional information, use the right tool, explain the result, and provide useful next steps.

Another challenge is reliability.

A false warning can cause unnecessary panic, while falsely telling someone that a scam is safe could be much more serious. Because of this, we are focusing on explanations, evidence, and uncertainty rather than simply producing a "safe" or "scam" label.

Since this is also my first hackathon, learning how to turn an idea into a working product while understanding MCP, conversational AI, and the Alexa+ requirements has been a major part of the challenge.

Accomplishments that we're proud of

We are proud that we started with a real-world problem instead of starting with a technology and looking for somewhere to use it.

We also wanted the project to demonstrate that an AI assistant can be more useful when it has access to specialized tools and can reason about when those tools are needed.

Another thing we are proud of is the focus on responsible AI behavior. ScamShield is designed to explain its reasoning and acknowledge uncertainty instead of presenting every AI-generated answer as a fact.

Most importantly, this project represents our first step into building a complete AI application for a real-world use case.

What we learned

This project has taught us that building an AI application is much more than connecting an LLM to a chat interface.

We learned about conversational AI, MCP, tool-based workflows, API integration, backend development, and the importance of designing for reliability.

We also learned that a good AI product needs to consider what happens when the AI is uncertain or wrong, not just what happens when everything works perfectly.

As a first-time hackathon participant, I also learned how important it is to start with a focused problem, build a working version first, and then improve it rather than trying to build everything at once.

What's next for ScamShield AI — A Conversational Safety Assistant

This hackathon is only the beginning.

The next step is to make ScamShield more capable while keeping it simple and trustworthy.

Some of the areas we would like to explore next are:

• Better analysis of suspicious messages and links.
• More reliable verification using trusted sources.
• Support for additional Indian languages.
• Voice-first interactions.
• Personalized safety guidance.
• Better protection against phishing and impersonation attempts.
• Additional tools that can help users take safe actions after discovering a potential scam.

Our long-term vision is to make ScamShield feel less like a traditional security tool and more like a trusted conversational companion that people can turn to whenever something online doesn't feel right.

Think before you click. Ask ScamShield first.

Built With

• agent
• ai
• alexa+
• amazon-web-services
• artificial
• assistant
• conversational
• cybersecurity
• detection
• fastapi
• generative
• intelligence
• language
• learning
• machine
• mcp
• natural
• phishing
• processing
• python
• scam
• voice