Weekly Sync - 7 July 2026

Attendees

• Swapnil Dutta
• Shaheem Azmal M MD
• Kaushlendra Pratap

Summary

• Discussed integrating the Nirjas ML classifier using the model2vec and potion-base-32m model (CPU-optimised inference considerations).
• Referred to Safaa's codebase for implementation patterns and examples.
• Raised a PR for the same to the Nirjas repository: https://github.com/fossology/Nirjas/pull/78 (built on PRs 76 → 77 → 78).

Notes

• Model choice aligns with prior benchmarking (model2vec family): small footprint (~30 MB) and high throughput on CPU.
• Emphasised data-quality and deployment considerations (quantisation, batching, and fallback routes).