Previous experience · Pic-Time
AI Album Designer
A real-time feature that turns a photographer's full shoot into a finished album.
My role
AI Engineer · 04/2022 to 09/2026
Designed and built the AI Album Designer as part of the AI team: the multi-model pipeline (face recognition and clustering, automated photo selection, saliency-aware smart cropping) and the inference work that lets albums generate while the photographer waits. Also shipped face-based gallery access and keyword- or face-based photo search. It started as an internship project and grew into a core AI team and product pillar.
Delivered a 10x improvement in inference throughput in automated album design · dataset curation across 50,000+ galleries
Problem
Designing an album from a full shoot takes photographers hours. The feature does it in real time, while they wait.
Role
Designed and built the AI Album Designer as part of the AI team: the multi-model pipeline (face recognition and clustering, automated photo selection, saliency-aware smart cropping) and the inference work that lets albums generate while the photographer waits. Also shipped face-based gallery access and keyword- or face-based photo search. It started as an internship project and grew into a core AI team and product pillar.
Approach
A multi-model pipeline: face recognition and clustering, automated photo selection and saliency-aware smart cropping, with near-duplicate rejection, album-wide consistency and faces kept intact under tight crops. Multi-stage inference was optimised so albums generate on demand.
Stack
Python · FastAPI · Docker · Kubernetes · Multimodal transformers · Embeddings · vector search.
Results
Delivered a 10x improvement in inference throughput in automated album design by adapting transformer architectures to production constraints.
Built and maintained the end-to-end training pipeline for proprietary multimodal models, covering dataset curation across 50,000+ galleries and containerized deployment via Docker and Kubernetes.