A developer open-sourced a local video retrieval engine built on a fine-tuned EVA-02 model after a 7-day sprint, designed to search large unstructured video portals by visual poses, clothing details, camera angles and motion cadence while ignoring uploader metadata at search time and running in minimal VRAM. The project targets queries the developer says CLIP still handles poorly.
TL;DR: After a manic 7-day sprint, I open-sourced a local video retrieval engine built to search massive unstructured video portals by visual poses, clothing details, camera angles, and motion cadence—ignoring up metadata at search time and running in minimal VRAM. GitHub: https://github.com/q
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.