Ilias Haddad tackled a personal archive of 2,207 GoPro videos, each recording moments from his cycling adventures. To sift through the footage, he built a local indexing pipeline on his Apple M1 Max. The system uses open‑source machine‑learning models to flag highlights automatically across multiple days of rides captured daily.

The indexing engine processes 668.68 GB of video, yielding 15 hours of searchable metadata. Haddad’s workflow streams the identified clips straight into DaVinci Resolve, where he assembles the final edit without manual review and significant efficiency.

Running the pipeline locally sidesteps cloud costs and privacy concerns common to commercial video‑analysis services. By leveraging the M1 Max’s Neural Engine, the project achieves near real‑time inference, keeping processing times under a minute per hour of footage. This demonstrates the feasibility of on‑device AI for media workflows.

Haddad’s system showcases how hobbyists can build powerful media pipelines with consumer hardware. The open‑source stack—comprising TensorFlow Lite, FFmpeg, and custom scripting—remains fully reproducible. Anyone with an M1 Mac can replicate the setup, streamlining video curation for sports, travel, or archival projects and daily creativity.