Terms and Conditions Analyzer

Nobody reads the fine print. LEGAL FILTER uses Gemini AI to scan 50-page legal walls in seconds, instantly exposing how they sell your data and track your IP. Real consent, zero legalese.

Comment

ClearPrint: The Privacy Sentinel The Inspiration Frustrated by "legal armor," I built ClearPrint to transform 50-page Terms of Service into transparent data, treating legal jargon as a cybersecurity threat to be decoded. How I Built It

  • The Brain: Gemini 1.5 Pro to digest massive legal files.
  • The Engine: Python (BeautifulSoup, Requests) for automated legal text scraping.
  • The Interface: A minimalist Streamlit dashboard for aesthetic, split-view audits. The Challenges
  • Obfuscation: Tuning prompts to bypass vague terms like "trusted partners."
  • Scraping: Navigating dynamic JS-heavy sites to find hidden legal links.
  • Logic: Calculating a Risk Score R based on clause weights w and counts c:

The Lesson AI is the ultimate equalizer, turning complex "walls of text" into instant, informed consent.

Built With

Share this project:

Updates

posted an update —

Project Update: ClearPrint — The Privacy Sentinel It’s been an incredible journey evolving ClearPrint from a simple Python script into a robust "Privacy Sentinel." As we move through 2026, the goal remains the same: Total Transparency. What’s New?

  • Automatic Web-Scraping 2.0: No more hunting for the "Privacy" link in the footer. You can now just drop a homepage URL, and our engine automatically identifies, crawls, and audits the legal sub-pages.
  • Gemini 1.5 Pro (2M Context) Integration: We’ve officially moved to the 2-million-token context window. This means you can now audit massive, multi-document enterprise agreements without losing a single clause.
  • Visual Risk Dashboard: We’ve replaced the "wall of text" results with an aesthetic, card-based UI that highlights Data Monetization, IP Rights, and Technical Tracking at a glance. Inside the Code We’ve refined our "Red Flag" detection logic using a weighted heuristic to calculate the Risk Score R: Where w_i is the severity of the clause (e.g., 10 for "Data Selling"), c_i is the count, and T is the total token count. Challenges Overcome The biggest hurdle lately has been Legal Evasion. Modern T&Cs are using increasingly vague language (e.g., "trusted ecosystem partners") to mask data-sharing. We’ve countered this by tuning our System Prompt to act as a "Phish-Hunter," specifically looking for technical identifiers like IP tracking and browser fingerprinting that marketing-speak can't hide. Screenshot Sneak-Peek > [Image: A sleek, deep-charcoal dashboard showing a "High Risk" alert for a popular social media site, with specific clauses highlighted in amber.] > Thank you to everyone who has been testing the beta! Your feedback on the "Human Translation" toggle has been invaluable.

Log in or sign up for Devpost to join the conversation.