**The Problem:** Many operators of nuclear reactors are not trained in reactor physics, some only have a high school diploma / GED, and as such, they are unable to make informed decisions about certain data-driven areas of reactor operations. This presents a major issue during safety-critical operations.

**The Solution:** A natural language interface powered by an LLM that allows operators to easily query and understand the state of reactor components. This system interprets complex data generated by a custom simulation of the Illinois Micro Modular Reactor (built on export-controlled specifications using the MOOSE framework).

**The Traction:** The system was presented to Idaho National Laboratory, and the Illinois Micro-Nuclear Reactor team, where it was validated and implemented as a proof of concept.

**The Tech:** I created the simulation system using MOOSE, a physics-based simulation framework written in C++. We used a local LLM, finetuned LLAMA 2 70B, and a local Vector Database to store important data the LLM needed to know about our specific nuclear reactor.