Pipeline

01 — Ingest LIVE

Raw footage drops into the pipeline. FFmpeg probes each file for codec, resolution, duration, and audio channel layout. Metadata is extracted and normalized into a structured manifest.

Tool: ffmpeg-python

Output: JSON manifest per clip

Formats: MP4, MOV, MKV, WAV

02 — Transcribe LIVE

GPU-accelerated speech-to-text via faster-whisper (CTranslate2 backend). Each clip produces timestamped segments with word-level alignment. Full-text search index built automatically.

Engine: faster-whisper 1.2.1

GPU: RTX A6000 float16

Output: 148 segments across 20 clips

03 — Analyze LIVE

NLP-based chapter boundary detection. Editorial scoring across four weighted dimensions: dialogue density (0.35), pacing (0.25), content quality (0.20), and technical quality (0.20). Dialogue matching groups content across takes.

Chapters: 27 detected

Scores: 3.0–7.3 (mean 4.9)

Match groups: 175

04 — Catalogue LIVE

All analyzed content is indexed into a searchable catalogue. Thumbnails auto-extracted at chapter midpoints. Clips sortable by score, duration, chapter, and content type.

Entries: 20 clips catalogued

Thumbnails: 27 auto-extracted PNGs

Search: Full-text NLP queries

05 — Rough Cut NEXT

Score-driven clip assembly. Top-scoring chapters are auto-sequenced into a rough timeline. Silence gaps are trimmed. The result is an EDL (Edit Decision List) ready for import into Resolve.

Input: Scored catalogue

Output: EDL + timeline JSON

06 — Export PLANNED

DaVinci Resolve Studio API integration for programmatic timeline building. Multi-format export targeting YouTube (16:9), TikTok/Shorts (9:16), Instagram (1:1), and archive-quality ProRes.

Integration: Resolve Studio 20.3.1.6

Wrapper: pybmd Python API

07 — Distribute PLANNED

Post-publish analytics from AnaBot feed back into the pipeline. Engagement metrics inform the next edit cycle. Performance data shapes future rough cut selection criteria.

Feedback: AnaBot analytics

Output: Next-cycle edit brief

Stages

Processing Flow