Skip to content

AI · Mobile & API

ClearView Caption

Real-time image captioning and scene summaries for blind and low-vision users — sub-2s latency on mid-range phones, WCAG-aligned UX.

Access Media Cooperative · 11 weeks · 2025

ClearView Caption software dashboard with scene preview, generated caption, and vision pipeline metrics.

Challenge

The cooperative’s members relied on volunteers to describe images over phone calls — slow, inconsistent, and unavailable after hours. They needed an on-device-friendly app that could caption gallery photos and live camera frames without sending raw pixels to unknown third parties.

Approach

We built a React Native app with a tiered inference path: on-device classification for common objects, then a hosted vision-language model with PII-safe cropping and audit logs. Captions are post-edited for reading level, cached at the edge, and read aloud via native TTS with adjustable speed.

Timeline

  1. 01

    Accessibility research

    Week 1–2

    Sessions with 14 blind and low-vision testers. Defined caption tone, error states, and offline fallbacks.

    • Research synthesis
    • UX flows
    • Safety & privacy model
  2. 02

    Vision pipeline MVP

    Week 3–5

    Camera capture, gallery import, caption API, and on-device preprocessing. Latency budget enforced in CI.

    • Caption API
    • Mobile alpha
    • Latency dashboards
  3. 03

    Beta & localization

    Week 6–8

    English and German caption styles, haptic cues, and high-contrast UI. 200-user closed beta.

    • TestFlight / Play beta
    • Localization pack
    • Support playbook
  4. 04

    Production launch

    Week 9–11

    Staged rollout with model versioning, human review queue for edge cases, and handover runbooks.

    • App Store release
    • Model ops guide
    • Monitoring

Stack

  • Mobile

    React Native, Expo, Native TTS, Core Haptics

  • AI / Vision

    GPT-4o Vision, Custom prompt guardrails, Embedding cache

  • Backend

    Node.js, Redis, PostgreSQL, S3 (encrypted blobs)

  • Infra

    AWS, CloudFront, GitHub Actions, Sentry

“Our site went from a 4.2-second load to 1.1 seconds. Demo requests doubled the same month we launched — and our researchers can now publish a paper summary in five minutes instead of filing a ticket. That’s the whole review, honestly.”

Priya Raman, Head of Growth — Lumen Labs

Want similar results?

Get a free estimate and timeline within 48 hours.

Get a free estimate