Garg worked on the PriorLabs/TabPFN repository, focusing on improving the robustness of text feature preprocessing in machine learning pipelines. He addressed a critical issue with missing values in text data by developing the _process_text_na_dataframe utility, which fills NA entries with a placeholder and ensures proper encoding for both classifier and regressor models. Using Python and SQL, Garg implemented comprehensive end-to-end tests to validate that NaNs in text inputs are handled correctly during training and inference. This targeted bug fix enhanced data quality and model stability, supporting more reliable production deployments and safer text-based modeling with TabPFN.