GROWL 01 / 08
Available to Build For You

GROWL

A production intelligence system I designed and built from scratch -- and a blueprint for what it could look like inside your organization.

Offline Local Proxying • Cloud Serverless Enrichment • OTIO Export
GROWL 02 / 08

The Creative Black Hole

  • Offline B-Roll Trap High-resolution production files sit inactive on local hard drives.
  • The Search Penalty Editors waste up to 30% of their working hours manually scrub-searching for clips.
  • The Bandwidth Bottleneck Uploading terabytes of raw camera footage for AI cloud tagging is slow, expensive, and impractical.

The Current Reality

Creative agencies and production teams are drowning in media but starving for efficiency. Valuable footage remains buried because cataloging is subjective and manually exhausting.

"We spent 4 hours looking for an alternate camera angle."
GROWL 03 / 08

What I Built: Decoupled AI Ingestion

I designed and implemented this architecture end-to-end. Growl bypasses internet bottlenecks by separating local proxy rendering from cloud-scale AI analysis.

1. Growl Prep (Local)

Lightweight macOS desktop app probes RAW clips and transcodes clean H.264 review proxies locally.

0% CLOUD UPLOAD COST
→

2. Proxy Upload (1% Bandwidth)

Only the small review proxies are sent to Cloudflare R2. Original RAW source files remain secure on local drives.

ULTRA-FAST TRANSFERS
→

3. Cloud GPU Pipeline

Serverless GPU workers ingest proxies and execute heavy deep-learning computer vision and audio modeling.

100% AUTOMATED
GROWL 04 / 08

The 12-Step Enrichment Engine I Built

Vision Steps 4-7

FFmpeg scdet scene detection, black frame registers, and optical-flow camera movement analysis.

Deep Object Step 8

YOLOv8 & Pose object detection. Face verification, spatial coordinates, and bounding box coverage.

CLIP Embedding Step 8.5

openai/clip-vit-base semantic vector embeddings mapped to each keyframe for similarity searches.

Cognitive Description Step 9

Gemini Vision multimodal descriptions, narrative summaries, and shot classifications.

WhisperX OCR Steps 10-11

Word-level audio transcript, speaker diarization, and on-screen text chyrons/slide OCR.

Acoustic Energy Step 12

Librosa tempo, beat trackers, speaker segment RMS energy, and music/speech/silence filters.

GROWL 05 / 08

Instant Semantic Search

I built a working semantic search layer that runs zero-API-cost local queries combining dialogue text and vector CLIP model matching over keyframe embeddings. Here's a live demo.

  • Try the Live Demo: Type keywords like "sunset", "dogs", or "innovation" into the console!
🔍
🐕 Sunset Run
Beach B-Roll

02.14s - 05.80s 94% Match

🗣️ Speaker Presentation
Speaker: "...innovation is key..."

45.10s - 49.30s 89% Match

GROWL 06 / 08

Conformed OTIO Timelines

  • Built for Newsrooms and Large Media OrgsFaster turnaround on archival requests, editors spending time editing instead of hunting for clips.
  • AI Script Advisor Growl reads a natural-language brief, drafts a three-column script, and gathers fitting clips.
  • Handles-Aware R2 Trimming The conform engine streams and cuts exact ranges with head/tail handle extensions on R2.
  • Pixar OpenTimelineIO Assembly Exports conformed timeline arrays straight into standard .otio file payloads.
CONFORMED OTIO TARGET:
VIDEO 1
Clip_01 (Fred_PSA) +2s handles
Clip_02 (Sunset) +2s handles
AUDIO 1
Speech segment track (AAC)
• Import directly to Premiere, Resolve, or Final Cut. Relinks to local high-res RAW instantly.
GROWL 07 / 08

What I Bring

What Makes This Different

Most AI developers don't understand production workflows. I do -- as a filmmaker who has felt this pain directly. Growl is proof of concept, technical portfolio, and a system ready to be adapted for your infrastructure.

  • Working PrototypeNot a concept -- a functional 12-step enrichment pipeline running on real footage.
  • Production Domain KnowledgeUnderstanding of editorial workflows, NLE standards, and how footage actually moves through a production.
  • Ready to AdaptThe architecture is designed to be modular. It can be scoped to your stack, your storage, and your team's workflow.
GROWL 08 / 08
Let's Talk

Adam Behrmann

Filmmaker and AI systems builder. I built Growl to solve a problem I lived firsthand -- and I'm looking for the right organization to bring this capability to at scale.

Production Background • AI Pipeline Engineering • Editorial Workflow Design
Use LEFT / RIGHT arrow keys or SPACEBAR to navigate