Creating AI Software Prototypes to Support Academic Research

Motivation and context

Film- and media-studies research often requires locating specific moments in large audiovisual corpora: recurring motifs (e.g., “rain-soaked street at night”), visual styles (e.g., “static wide shot, pastel palette”), themes (e.g., “bureaucracy and waiting rooms”), or objects/events (e.g., “train entering a tunnel”). Traditional metadata and keyword search rarely captures this well.

This project will build a small prototype that enables semantic search over video using natural-language queries and optional visual queries (image upload). The prototype aims to support exploratory research workflows (close reading, motif tracking, scene comparison) and to serve as a foundation for future, larger-scale tooling.

A strong reference point is aninibread/ghibli-search, which demonstrates semantic search over movie stills using Cloudflare Workers + R2 + AI Search. https://ghibli-search.anini.workers.dev/