Tools built to support an independent feature film made without industry backing. Its behind-the-scenes footage is part of the film's story, which means the production itself generates a large, continuously growing body of raw media that needs to become searchable and editable rather than sitting unused on disk.
This repo holds the media-evidence retrieval pipeline that grew out of that need. It turns raw production footage into searchable, structured artifacts:
- ASR / transcription
- speaker- and entity-level enrichment (faces, objects, shot metadata, character clustering)
- vector embeddings + a search API over the resulting corpus
- a synthetic-query layer for natural-language lookups against the archive
It's tooling built for a real production, published as it's built rather than as a finished product. Expect rough edges and active change.
See MASTER.md for the canonical spec (stage responsibilities, invariants, execution rules) and
PIPELINE_WRITE_AND_INTERACTION_MAP.md / METADATA_TAGS_AND_FLAGS_CONTRACT.md for the artifact
contracts.
Development and processing for this pipeline is done in collaboration with Runpod, who provide GPU credits.
Active development, in step with an ongoing film production. Interfaces and stage boundaries may change without notice.
Code in this repository that is our own work is released under the MIT License, free for any use, including commercial, with attribution.
This pipeline depends on and integrates with third-party libraries, models, and tools, each under their own license terms (some permissive, some more restrictive). Those third-party components are not relicensed by this repository. See their own upstream projects for terms. A third-party notice file will be added here as that audit completes.