Conversation risk analyzer

• 

  Safe dialog

• 

  High risk conversation with detailed red flags

Inspiration

Online harassment and stalking affect millions of people daily. Many victims struggle to identify whether someone's behavior is genuinely threatening or just uncomfortable. I wanted to build a tool that helps people make that distinction objectively.

What it does

Conversation Risk Analyzer lets you upload a screenshot or paste text from any conversation. It analyzes the messages for red flags like manipulation, threats, gaslighting, and escalation patterns — then returns a risk score (0-100) with specific explanations.

How we built it

• Google Gemini API for multimodal conversation analysis
• Streamlit for the web interface
• Python for backend logic
• Deployed on Streamlit Cloud

Challenges we ran into

• Getting Gemini to return consistent structured JSON output
• Handling both image screenshots and raw text input
• Securing API keys properly for cloud deployment

Accomplishments that we're proud of

• Works with both screenshots and text in real time
• Returns actionable recommendations, not just a score
• Deployed as a publicly accessible web app

What we learned

Prompt engineering is crucial — the difference between a vague and a precise prompt completely changes output quality.

What's next for Conversation Risk Analyzer

• Support for multiple languages
• Chat history tracking across sessions
• Integration with reporting resources (hotlines, legal aid)

Built With

• gemini
• python
• streamlit