Scam Spotter

• 

  we have try out examples of actual scam messages and safe messages

• 

  it shows the message is scam and also guide you what to do next

• 

  we entered a safe message ,and it says no scam

• 

  and when i upload a screenshot of an unknown login page, it shows its risky

• 

  why its risky and what we should do now is already mentioned here

Inspiration

Scams reach every community, and the people hit hardest are often elderly or less tech-savvy users. When a suspicious "your account is blocked" message arrives, they usually have no quick, judgment-free way to check it.

What it does

Scam Spotter lets anyone paste a suspicious message or upload a screenshot (SMS, WhatsApp, email) and instantly get:

• a clear verdict: Scam / Risky / Safe
• a confidence score
• up to three red flags
• a plain-language explanation
• simple next steps (e.g. don't click the link, contact your bank directly)

It doesn't panic about everything: in testing, a fake Wells Fargo alert came back as Scam, an unofficial login page screenshot as Risky, and a normal family dinner message as Safe.

How we built it

• Python + Streamlit for the web app and UI
• Google Gemini API (via the google-genai SDK) for analysis, with a system prompt that makes the model act as a fraud-detection expert
• Structured JSON output validated with Pydantic, so the UI always gets a consistent verdict, score, flags, explanation and actions
• Image input: screenshots are sent directly to Gemini, which reads the text and visual cues
• Deployed on Streamlit Community Cloud, with API keys kept in secrets and never committed to GitHub
• Built with the help of an AI coding agent (Google Antigravity) for code generation, with testing and iteration done by hand

Challenges we ran into

• Calibrating verdicts: getting the app to separate "Risky" from "Scam" and not flag normal messages took prompt tuning and testing with different message types.
• Keeping secrets safe: I had to keep the API key out of the repo (.gitignore, .env.example) and configure it separately on the cloud deployment.
• Deployment hiccups: my first deploy ran without the API key and failed, so I learned how Streamlit secrets work and fixed it.
• First time with Git and agents: setting up Git, pushing to GitHub and working with an AI coding agent were new to me.

Accomplishments that we're proud of

A working, publicly deployed app that handles both text and screenshots, and gives explanations that a non-technical person can actually understand.

What we learned

How to build and deploy an LLM-powered app end to end, how to get reliable structured output from a model, and why security basics (secrets, .gitignore) matter from day one.

What's next for Scam Spotter

• Link reputation checks against known-bad URL databases
• Multi-language support so more communities can use it
• Voice input for people who find typing difficult
• A WhatsApp bot, so people can forward a suspicious message and get an answer in the same chat

Disclaimer: Scam Spotter is informational only. AI can make mistakes, so always verify with your bank or an official source.

Built With

• geminiapi
• git
• github
• json
• python
• sdk
• streamlit