Worked on enhancing data privacy in the BerriAI/litellm repository by improving the PII anonymization workflow. Addressed a bug in the mapping of PII tokens to original text by refining text-position handling and stabilizing token numbering, which reduced the risk of data leakage and improved the reliability of anonymized outputs. Strengthened the integration with the Presidio endpoint to ensure accurate token mapping and prevent garbled output. Focused on backend development using Python, with an emphasis on API integration and unit testing. The work included removing obsolete code, enforcing deterministic token order, and adding regression tests to maintain code quality.