A RAG-Enabled Large Language Model with Knowledge Graph Integration for Live Telemetry-Driven AI Assistance of Lunar Payload Operations

We present a conversational AI assistant developed for two MIT payloads on the Intuitive Machines-2 (IM-2) lunar mission: the AstroAnt micro-rover and a Kinect-based depth camera deployed on the Nova-C lander. To support both mission control and public engagement, we built an interactive Unreal Engine dashboard that visualizes live telemetry, recent commands, and system status integrated into a 3D visualization dashboard. At its core, the dashboard's AI assistant integrates a Large Language Model (LLM) augmented with Retrieval-Augmented Generation (RAG) and a continuously updated Knowledge Graph (KG). This hybrid architecture enables users to query real-time telemetry, such as temperatures, voltages, or sensor trends, while also retrieving contextual information from mission documents and public sources. The KG ensures structured grounding of data streams, while the vector database supports semantic retrieval across heterogeneous mission assets. During deployment, users could interact with the assistant to explore payload operations in natural language, demonstrating reliable retrieval of live and historical mission data. This paper describes the system architecture, including telemetry integration, knowledge graph schema, and query orchestration with LangGraph. We further discuss limitations of semantic retrieval for space mission data and outline future directions in multimodal embeddings and trust calibration. The results highlight how RAG-enabled LLMs can enhance operational decision-making and simultaneously engage the public with transparent, verifiable insights into lunar exploration.