Fix: Preserving user_email in audio job status to prevent 403 errors

Fix: Final bulletproof video pipeline with whisper and full fallbacks

Fix: Added sentencepiece and improved model patching with fail-safes

Fix: Moondream2 compatibility and added fail-safe for reasoning engine

Fix: Preserving user_email in video job status to prevent 403 errors

fix: use safe get() for DB save to prevent KeyError

fix: update protobuf version to avoid pip conflict

fix: install reportlab in Dockerfile

fix: add reportlab to requirements

clean backend deploy

Fix: Add spaCy and pre-download en_core_web_sm model for Text Lab

Optimization: Move email-validator to end of Dockerfile to preserve heavy ML layer cache

Fix: Add missing email-validator for Pydantic authentication schema

Build Fix: Granular isolation of final dependencies

Fix: Add CPU torchvision to prevent pyannote from downloading GPU torch and hitting disk limits

Fix: Restore full build with correct libgl1 package

Debug: Replace libgl1-mesa-glx with libgl1

Debug: Test apt-get in minimal Dockerfile

Debug: Minimal Dockerfile test

Debug: Remove libmagic and python-magic to test apt-get

Debug: Full disaggregation to isolate failing package

Fix: Multi-stage pip installs with full fault isolation

Critical Fix: Slim requirements and clean Dockerfile to fix HF disk space issue

Build Fix: Split requirements into individual RUN commands for precise caching and debugging

Fix syntax in Dockerfile and requirements

Build Fix: Add build-essential and cmake for C extensions

Build Fix: Remove unused retina-face to prevent build timeout

Build Fix: Remove torch from requirements to avoid CPU build conflict and add libmagic1

Optimization: Split pip installs for CPU-only build stability

Build Fix: Explicitly include requirements.txt

Fix: Add app_port to HF metadata

Fix: Use valid Hugging Face metadata colors

Fix: Add Hugging Face Space configuration metadata

Initial Deploy: FakeShield Backend v2.0 (Sovereign Vanguard)