Multimodal + Rag + Inference

Build a bridge connecting geospatial foundation models (like PDFM) directly into LLM latent spaces for reasoning tasks. Enable LLMs to 'see' geospatial data and reason over population dynamics.