DEPLOYMENT.md

BAKO-AI Deployment Guide

This guide explains how to manage and deploy the BAKO-AI analysis pipeline to Hugging Face.

🏗 Architecture Overview

• Application: Hosted on Hugging Face Spaces (icanedit2/BakoAI).
• Models: Hosted on a dedicated Hugging Face Model Repository (icanedit2/bakoai-models).
• Storage/DB: Managed via Supabase.

🚀 Deployment Workflow

1. Push Code Changes

To push local changes from the back-end folder to the Hugging Face Space root:

# From the root directory:
git subtree split --prefix back-end -b hf-production-build
git push hf hf-production-build:main --force
git branch -D hf-production-build

🧠 2. Managing Models

Models are stored in the icanedit2/bakoai-models repository to prevent git push from deleting them.

• To add a new model, upload it to the bakoai-models repo.
• Update download_models.py in the Space repo to include the new filename.
• The build process will automatically download models into /home/user/app/models/.

🔐 3. Security (Hugging Face Secrets)

Ensure the following Secrets are set in your Space Settings:

• HF_TOKEN: Required for downloading models from private repositories.
• SUPABASE_URL: Your Supabase project URL.
• SUPABASE_KEY: Your Supabase anonymous/service key.
• JWT_SECRET: For authentication.

🎬 Video Playback Optimization

All annotated videos are processed through FFmpeg with the following settings for maximum browser compatibility:

• Codec: libx264
• Pixel Format: yuv420p
• Dimensions: Forced to even values (H.264 requirement)
• Streaming: +faststart enabled for instant playback.