Building Glopaw
Glopaw is a community app for wildlife watchers. You photograph an animal, a vision model identifies the species with a confidence score, and you can share the sighting to a feed where other people corroborate or correct it.
Why I built it
Wildlife ID apps mostly treat the photo as a lookup — you get a species name and that's it. The interesting part to me was the community layer: a vision model is confident a lot of the time and wrong some of the time, and the way you fix that is to let people who know the local fauna weigh in. So I built the identification and the social feed as one thing, not the feed as an afterthought bolted onto a classifier.
What it does
Mobile app (React Native)
Cross-platform on iOS and Android. The capture screen is tuned for wildlife — shooting at distance, dealing with movement — with optional on-device cleanup before upload. There's a feed with chronological and algorithmic views, a profile with your observation history, and offline capture so you can shoot in the field with no signal and let it sync when you're back in range.
Backend (Node.js + Drizzle ORM)
The Node backend runs auth (with privacy controls over location sharing, since "where I photographed this animal" is sensitive data), a PostgreSQL database for accounts, image metadata, and identification history, and a REST API between the app and server. There's a moderation path that mixes automated checks with human review to keep the feed clean.
Identification pipeline
Images land in AWS S3, then go to a Llama vision model for species identification across taxonomic categories. Location, time, and environmental data get attached to sharpen the result, and every identification comes back with a confidence score plus alternatives when the model isn't sure — I'd rather show uncertainty than a confident wrong answer. Edge cases and rare species route to community experts for validation.
The conservation angle
This kind of data is useful beyond the app:
- Distribution mapping from anonymized, aggregated sightings.
- Rare-species alerts when a threatened animal turns up.
- Species context — conservation status and habitat info alongside the ID.
- Community science campaigns to feed structured observations into real research.
What makes it different
Most wildlife ID apps are a classifier with a thin social skin. Glopaw is community-first: the feed and the corroboration loop are core, the vision model is tuned for wildlife rather than general images, and location data is handled with granular privacy controls because publishing exact coordinates of rare animals is a genuinely bad idea.
What's next
- Behavior recognition, not just species.
- Ecological relationships between observed species and their environment.
- Temporal analysis of seasonal sighting patterns by region.
- AR overlays for educational context during live viewing.
- A research API so institutions can use the observation data.