AccessLens

Turn one venue photo into evidence-linked accessibility barriers, prioritized fixes, and measurable improvements with Google Gemini.

Comment

Inspiration

Many small shops, clinics, cafés, colleges, and community spaces want to become more accessible but do not know what to inspect or fix first. Generic checklists lack context, while a professional accessibility audit may not always be immediately available.

We built AccessLens to make that first step visible and actionable.

What it does

AccessLens turns one venue photograph into an evidence-linked accessibility action plan.

Users can upload an entrance photo and provide basic venue context. Google Gemini then:

  • Identifies visible accessibility barriers
  • Highlights evidence directly on the photograph
  • Explains who may be affected
  • Assigns severity and confidence
  • Recommends prioritized improvements
  • Estimates cost, effort, and score impact
  • Generates English and Hindi summaries

Users can complete recommendations and immediately see the accessibility score improve.

AccessLens is an AI-assisted screening tool, not a legal compliance certification. Structural measurements and final decisions should be verified by a qualified professional.

How we built it

AccessLens uses React and Vite for the responsive frontend, with Node.js and Express for the backend API.

The backend sends the venue image and context to Google Gemini 3.5 Flash-Lite using Google's official GenAI SDK. Gemini returns structured JSON containing evidence-grounded barriers, image regions, confidence values, affected users, and prioritized recommendations.

MongoDB and Mongoose are supported for persistent audits, with an in-memory store available for the hackathon prototype. The application also includes image validation, analytics events, bilingual output, responsive layouts, and an explicitly labeled offline fallback.

Challenges we ran into

The largest challenge was using AI honestly. A single photograph cannot prove exact dimensions or certify legal compliance.

We addressed this by:

  • Requiring visible evidence for every finding
  • Showing confidence values
  • Preventing unsupported measurement claims
  • Treating text inside uploaded images as untrusted content
  • Using a strict structured-output schema
  • Adding a human-verification checkpoint
  • Clearly identifying live Gemini and offline-demo results

We also migrated to the latest Gemini Interactions API and selected Gemini 3.5 Flash-Lite for reliable free-tier multimodal performance.

Accomplishments that we're proud of

  • Built a complete photo-to-action workflow
  • Successfully integrated live Gemini multimodal analysis
  • Added interactive evidence annotations
  • Created measurable score updates when actions are completed
  • Added English and Hindi summaries
  • Designed loading, error, empty, onboarding, and offline states
  • Tested desktop and mobile layouts
  • Passed all backend integration tests
  • Created a deterministic fallback for internet or API failures

What we learned

We learned that trustworthy AI is not only about producing an answer. Users need to understand why the model made a claim, how confident it is, and what still requires human verification.

We also learned that a focused end-to-end workflow is more valuable than adding unrelated AI features.

What's next for AccessLens

Next, we plan to:

  • Work with disabled users and accessibility professionals
  • Calibrate the scoring rubric against expert assessments
  • Support multi-image venue audits
  • Add Marathi and additional regional languages
  • Generate professional audit-ready PDF reports
  • Add secure image storage, authentication, and deletion controls
  • Build an expert-review workflow for structural recommendations

Built With

Share this project:

Updates

Submission history