Scam Spotter AI

• 

  Main UI

Inspiration

We've all seen it — "Buy this crypto NOW!" or "This one trick will make you rich by 20." Social media is filled with people offering investment advice as though they're professionals but half of them sound like walking cons. I wanted to design something simple that can help regular people know if what they're hearing is smart. or a scam.

What it does

The user pastes the financial advice he took from the internet guru, and AI analyzes the advice and gives it a confidence rate according to how likely this is a scam, the lower the confidence the higher the probability of the advice being a scam.

How we built it

I'm using Streamlit for the frontend (pretty and clean, does the trick). The real smarts of the project are the facebook/bart-large-mnli model by Hugging Face. They enable me to classify any financial quote into: Safe Advice Risky Advice Scam Advice User pastes a quote, clicks a button, and voilà — the AI gives its verdict. All hosted on Replit, because I like pain (tongue in cheek… kind of).

Challenges we ran into

The API sometimes bailed on me with vague error messages. That slowed things down but also forced me to write better error handling. Installing things on Replit was a challenge. Pip randomly didn’t work sometimes. Figuring out the right labels and prompts for the AI was harder than I expected. One word change could throw off the whole result. Also, making something look simple is harder than it sounds.

Accomplishments that we're proud of

Got a working AI-powered scam detector running in under a week — no training, no fancy setup, just clean logic and smart tools. Built a web app that actually solves a real problem — not just another to-do list. Managed to connect Streamlit + Hugging Face API + Replit, which wasn’t always smooth, but it works. Kept it simple enough that anyone can use it — no sign-ups, no technical barrier, just paste and check. Created a project that could genuinely help people avoid sketchy advice online.

What we learned

I finally got a solid grasp on how zero-shot machine learning works (you don’t need to train anything, it just gets it). Learned how to work with Hugging Face APIs, which is like giving your app instant AI powers. Got better at building real-time web apps using Streamlit, which is surprisingly smooth. Also realized that even if your app is basic, if it solves a real problem, people actually care.

What's next for Scam Spotter AI

Add a keyword-based scam warning layer — highlight common red flags like “guaranteed returns,” “DM for crypto,” etc. Build a finfluencer score system — let users rate advice over time, and build credibility profiles. Make it mobile-friendly — most people see this junk on their phones anyway. Try fine-tuning a model using real scam/risky/safe advice for better accuracy. Maybe turn this into a browser extension so users can scan advice directly from TikTok, Twitter, or YouTube.

Built With

• huggingface
• python
• streamlit