DataGemma RIG is a series of fine-tuned Gemma 2 models aimed at helping large language models (LLMs) access and integrate reliable public statistical data from Data Commons. This model employs a retrieval-based generation approach, annotating statistical data in responses through natural language queries to Data Commons' existing natural language interface. Trained on TPUv5e using JAX, DataGemma RIG is currently in the early stages and is intended primarily for academic and research purposes; it is not yet ready for commercial or public use.