BioLens

Setting Screen
Home Screen
Processing Picture
EXAMPLE of raisable organism
EXAMPLE of non-raisable organism

Inspiration

A pocket field guide for any living thing — so anyone with a phone can identify an organism and learn how (or whether) to care for it.

What it does

Point your camera at any plant, animal, or fungus and BioLens identifies it, then assembles a care guide with habitat, diet, size, lifespan, raisability, and warnings about endangered or protected species.

How I built it

SwiftUI iOS app that runs identification through the BioCLIP vision foundation model on a Hugging Face Gradio Space, then layers Wikipedia, Wikidata SPARQL, and an LLM extraction step (HF Inference + Qwen2.5) to build a structured care profile.

Challenges I ran into

BioCLIP has no hosted REST API — I reverse-engineered the Gradio queue protocol, handled /gradio_api/ path changes, parsed mixed-type SSE responses, and tuned accuracy.

Accomplishments that I'm proud of

I built a system that automatically converts Wikipedia's vast biological knowledge into structured species care profiles, enabling BioLens to scale to thousands of organisms without manual data entry.

What I learned

I learned that species identification is much harder than it seems—even strong vision models can struggle with similar-looking organisms, image quality, and uncommon species. Building a useful app required designing around uncertainty rather than assuming the AI was always right.

What's next for BioLens

Bundling a Core ML BioCLIP build for fully on-device identification, adding iNaturalist citizen-science contributions so scans help real biodiversity data, and surfacing IUCN conservation status as a hero badge to push the social-good angle further.