Dawn Chorus: An Offline Bird-Call Bingo Built on BirdNET and Gemma (hacktoberfest week 1 dev challenge)

This is an account of Dawn Chorus, a hackathon project submitted to the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass. Dawn Chorus is an offline bird-identification app that runs entirely on a laptop during a walking outing.

Upon opening the app, users are presented with a bingo card listing the 12 most likely bird species to be encountered in their area that week. The user then places their phone in airplane mode with the screen turned off, recording their walk with the built-in voice recorder. After returning home, the audio recording is uploaded to the app.

An open-source BirdNET model identifies the bird species present in the recorded audio based on their calls. Simultaneously, a locally running Gemma 3 model (4B) generates a short field-journal page describing the walk and its key observations. The journal concludes with a list of the bird species that were identified but not included on the bingo card, motivating the user to go out again tomorrow to seek out those additional birds.

The app emphasizes minimizing screen time during the bird-watching experience, aiming for the screen to be the shortest part of the activity. A "Grass Ratio" card tracks the time spent outdoors compared to the time spent interacting with the app's screen. Dawn Chorus is designed to provide a bird-watching experience that is accessible even in areas without internet connectivity.

The app works entirely offline as BirdNET identifies the birds from the audio recording on the user's laptop, and the language model's output is filtered to prevent false information being presented in the journal.