Skip to content

Zenieverse/SHEild

Repository files navigation

SHEild – AI Safety Planner

https://sheild-ai-safety-planner-764082783379.us-west1.run.app

Quickstart Instructions

Run the following command to set up and start the application:

npm install && npm run dev

Note: Ensure your GEMINI_API_KEY is configured in your environment secrets.

Decision Log

  • Architecture: Single Page Application (SPA) using React 19 and Vite for maximum speed and responsiveness.
  • AI Engine: Google Gemini 3 Flash chosen for its high-speed inference (sub-10s requirement) and robust structured output capabilities.
  • Styling: Tailwind CSS 4 used for rapid development of a high-contrast, accessible UI.
  • Anonymity: No backend database or authentication implemented to ensure 100% user anonymity and zero data persistence.
  • PDF Generation: Client-side generation using jsPDF and html2canvas to keep data within the user's browser context.

Evidence Log

  • Icons: Lucide React (ISC License).
  • Animations: Motion (MIT License).
  • PDF Engine: jsPDF (MIT License).
  • AI SDK: @google/genai (Apache-2.0).
  • Typography: Inter (Google Fonts, SIL Open Font License).

Risk Log

  • Issue Caught: AI initially suggested "calling for help" in situations where silence was safer (e.g., domestic tension).
  • Fix: Updated the Goose System Instruction to explicitly prioritize "discreet actions" and "text-based code words" over audible alerts in high-risk domestic contexts.
  • Issue Caught: PDF generation was cutting off text on mobile screens.
  • Fix: Implemented a responsive canvas scaling factor in html2canvas to ensure the full layout is captured regardless of the viewport size.

Known Limitations

  • AI advice is a guideline and not a replacement for professional emergency services.
  • Requires an active internet connection for Gemini API calls.

Accessibility & Reading Level

  • Color Contrast: WCAG AA compliant (verified via high-contrast mode).
  • Reading Level: Targeted at Grade 8 (simple, direct language used in UI and AI prompts).
  • Screen Readers: Semantic HTML and ARIA labels used throughout.

License

This project is licensed under the MIT License. See the LICENSE file for details.

About

AI-powered, trauma-informed safety planning tool built with Goose. 4-Line Problem Frame User: Women and marginalized individuals navigating unsafe environments Problem: Lack of fast, personalized safety planning tools Constraints: Privacy-first, no login, trauma-informed, safe AI output Success Test: User generates and downloads safety plan in unde

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors