Several hundred bird photos, automatically identified and sorted by species using an open-source AI model. All photos were taken by Ash, a novice photographer, using a Nikon P1000 and an iPhone.
This site was built as a personal project to organize 441 bird photos using AI. Photos were processed with BioCLIP, an open-source biodiversity identification model, which assigns a confidence score to each identification. Based on that confidence, photos are sorted automatically, flagged for a quick check, or set aside for manual review. Built on a self-hosted Linux server, using Nginx and Cloudflare Tunnel.
Each photo shows how confident the AI was in identifying it. 40%+ auto-sorted, high confidence. 25-39% flagged to confirm. below 25% uncertain, needs manual review.
The original goal was simple: use an open-source AI to identify birds from my own photos, tag each one with the species and a confidence score, and file it into the right folder automatically. That pipeline (BioCLIP + a sorting script) was built and works end-to-end.
What this page adds on top is a human review pass: a small admin tool I built lets me go through every photo the AI wasn't confident about, confirm the ones it got right, and correct the ones it got wrong. That review is what closes the loop, and what these numbers reflect.
Worth being upfront: human review isn't perfect either. I did my best to correctly identify birds the AI couldn't, but I'm not an ornithologist, and misidentifications are still possible in the "corrected" photos above — not just the AI-sorted ones.