Digital Armor — AI Scam Detector

• 

  interface of the project

Inspiration

Every day, thousands of people fall victim to online scams — fake job offers, phishing emails, and fraudulent messages. I wanted to create a tool that helps ordinary users detect scams instantly and build trust in digital platforms.

What it does

Digital Armor — AI Scam Detector uses OpenAI’s gpt-oss-20b model to analyze text from emails, job postings, or SMS messages. It returns:

A label: SCAM or SAFE

A confidence score

Reasons why it flagged the message (keywords, suspicious links, personal info requests)

How we built it

Frontend: Streamlit (simple, clean UI for pasting messages)

Backend AI: OpenAI’s gpt-oss-20b model via Hugging Face Inference API

Fallback checks: keyword & URL detection using Python regex

Deployment: Hugging Face Spaces (public demo)

Code repo: GitHub with README & setup instructions

Challenges we ran into

Ensuring JSON-only output from the model (sometimes it returns extra text).

Managing inference latency for large models.

Designing a safe UX that doesn’t give false confidence to users.

Accomplishments that we're proud of

Built a working prototype that detects scams in seconds.

Created a live demo accessible worldwide.

Designed a human-centered explanation system so users understand risks.

What we learned

How to integrate gpt-oss open-weight models into a web app.

The importance of combining AI + rule-based heuristics for safety.

How to package a project for hackathon submission (demo, repo, video).

What's next for Digital Armor — AI Scam Detector

Fine-tune gpt-oss on a larger scam dataset for higher accuracy.

Add multilingual support (detect scams in Hindi, Spanish, etc.).

Develop a browser plugin to scan phishing links in real time.

Built With

• github
• huggingfaceinferenceapi
• huggingfacespaces
• openaigpt-oss-20b
• python
• regex
• streamlit