All projects
AI / Tools · 2025Live
Photo Sorting AI
Turn thousands of event photos into a structured gallery in minutes.
Photo Sorting AI
Status
Live
Year
2025
Category
AI / Tools
Stack
Python, PyTorch, Vector DB
Case study
PythonPyTorchVector DBObject storageTypeScript
Overview
An AI-powered tool that organizes event photography by people, moments and quality — built for photographers and organizers.
Problem
Event photographers ship thousands of raw images and spend the next week sorting, tagging and culling. The work is tedious and high-stakes.
Idea
Use vision models to do the grunt work — group by person, surface the best shots, and let the photographer keep the final say.
Development process
- 01Profiled real photographer workflows to find the highest-leverage step.
- 02Compared multiple embedding models on real event datasets.
- 03Built a review UI that keeps the human in the loop without slowing them down.
Technical decisions
- 01Face embedding + clustering pipeline running in a queue worker.
- 02Quality scoring combining sharpness, exposure and composition heuristics.
- 03Object storage with signed URLs for secure delivery.
Challenges
- 01Clustering identities reliably across lighting and pose variance.
- 02Keeping inference cost per-event predictable.
Learnings
- 01Reviewable AI beats autonomous AI when stakes are high.
- 02Latency on the first 100 photos matters more than total throughput.
Future roadmap
- 01Client-facing gallery delivery
- 02Auto-album themes
- 03On-device pre-culling
Next project
Facout