timeglass

Walk through history, one snap at a time.

Comment

Inspiration: Panjim, Goa, is full of heritage buildings and stories most people walk past without noticing. Plaques and guidebooks are text-heavy and disconnected from the space itself. We wanted to flip that: point your phone at a place, and watch its history come alive right there, in real time. That's how Timeglass was born.

What it does: Snap Mode — take a photo of a heritage site, and on-device AI recognizes it instantly, no GPS or manual search needed. AR overlay & archival comparisons — a visual overlay blends with side-by-side archival photos, showing how the location has changed over decades. Whimsical voice narration — every location is narrated in a warm, slow, storybook-like voice instead of a flat robotic one.

How we built it: Timeglass is built with plain HTML, CSS, and JavaScript — no heavy framework. The Canvas API powers AR overlays and effects, on-device recognition identifies sites from a photo, and the Web Speech API drives narration, tuned with a slower rate and higher pitch (rate: 0.75, pitch: 1.3) for that fantasy-storyteller feel. It's deployed on Vercel for continuous deployment straight from main.

Challenges we ran into Deployment sync issues — Vercel occasionally showed a "stale" state where the live site lagged behind our latest commit, requiring careful cross-checking of commit hashes. Mobile quirks — speechSynthesis.getVoices() loads asynchronously on mobile, so we had to handle the voiceschanged event to avoid silent failures. Browser caching — some phones held onto old versions, making fixes look undeployed when they weren't. Scan mode — a continuous "walk and recognize" mode wasn't reliable in time, so we disabled it behind a feature flag rather than ship something broken. On-site testing — time constraints meant we couldn't visit every location for final testing, so we validated recognition using reference photos instead.

What we learned: How much AR/recognition work can be done with core browser APIs alone, how small parameter tweaks (like speech rate and pitch) can completely shift an app's feel, and that deployment pipelines can fail silently — verifying by commit hash beats assuming a push "just worked."

What's next for Timeglass? Expanding to more heritage sites and cities, building a fully working scan mode, and enabling community-contributed archival photos.

Built With

Share this project:

Updates

Submission history