Kaushi contributed to backend and DevOps engineering across multiple open-source repositories, focusing on reliability and maintainability. In stanfordnlp/dspy, Kaushi stabilized few-shot demo generation by ensuring metric thresholds were correctly propagated in unshuffled paths, adding regression tests in Python to prevent erroneous bootstrapping. For optuna/optuna, Kaushi enhanced type safety in visualization utilities by updating return type annotations for Matplotlib-based functions, clarifying API expectations when multiple axes are returned as numpy arrays. In bytedance/deer-flow, Kaushi improved Docker build security and reproducibility by introducing build argument support and resolving GPG signing issues, leveraging Dockerfile, YAML, and Python development expertise.