Worked on hardening the face-detection machine learning pipeline for the immich-app/immich repository, focusing on improving reliability when handling corrupted or empty images. Addressed a critical bug by introducing early validation of image dimensions using Python and OpenCV, ensuring that invalid inputs are caught before reaching the model. Centralized input validation at the API endpoint, which improved error messaging and prevented downstream failures. Enhanced type safety in the prediction path by resolving strict typing issues, contributing to maintainable backend code. Demonstrated skills in API development, backend engineering, and testing, resulting in a more stable image processing workflow for end users.