← sources

I made a real time fish identification app for scuba diving and snorkelling

source: reddit (reddit:ClaudeAI)published 2026-11-15 22:52ingested 2026-11-16 02:35

retained text

I recently made a real time fish identification app for scuba diving and snorkelling with the help of Claude and Codex! The app works completely offline using on-device models that I fine tuned, since reception doesn’t travel underwater. It’s called [FishSnapper](https://fishsnapper.com) and is [available on iPhone](https://apps.apple.com/us/app/fishsnapper-for-divers/id6814287970).

I had vibe coded a lot of stuff before, but this was my first time vibe training ML models. FishSnapper uses a pipeline of two models: a fine-tune of YOLO v9 for animal detection, and a fine-tune of BioCLIP 2.5 huge, which is a visual transformer, for classification. The data is largely from iNaturalist and includes some data I labelled for this project.

Something like this would’ve taken me way more time without AI coding tools. I could focus on research and user experience, and offloaded most of the operational work to Claude and Codex: assembling datasets, building data labelling UIs, renting GPUs, training models in the cloud, iterating on the iOS app, and more.

I initially had low expectations, but was quite surprised at how accurate it was. I wouldn’t bet on it getting the answer right every time, but it’s been reliable enough to make it easy to confirm and has already taught me a bunch of new species.

document fetched 2026-11-16 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xx98ed/i_made_a_real_time_fish_identification_app_for