fix(backend): restore text analysis on Render

[hrx01-dev]

Summary

• upgrade CPU-only PyTorch from 2.2.2 to 2.4.1
• align torchvision from 0.17.2 to 0.19.1
• keep the existing production text classifier and URL-analysis path unchanged

Root cause

Render was installing PyTorch 2.2.2 while the deployed Transformers runtime requires PyTorch >= 2.4. Because the backend groups its PyTorch/vision imports together, the incompatible runtime could cause those imports to fall back to None, leaving the text-analysis model unavailable at runtime.

Expected result

The Render container will install a compatible CPU-only PyTorch/torchvision pair, allowing the Hugging Face text-classification pipeline used by /api/analyze/text and URL content analysis to initialize normally.

Validation

• dependency versions are matched: torch 2.4.1 + torchvision 0.19.1
• Dockerfile already installs the requirements through the PyTorch CPU wheel index
• no API contract changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
veritas-ai	Error		Aug 27, 2026 8:04am