TRACEE: Traceable and Reliable AI for Computing Energy Efficiency

To address the DLA’s need for an energy use calculation tool, RedShred proposes to develop a new Traceable and Reliable AI for Computing Energy Efficiency (TRACEE) based on a combination of large language models (LLMs), generative AI (Gen AI) and intelligent prompt engineering. Specifically, the innovation in the ability of large language models (LLMs) to understand, calculate and reason within the mathematical domain via a more direct chain-of-thought (CoT) mechanism will enable the system to ingest scientific and engineering papers into LLMs to calculate energy use and emissions during manufacturing processes. As a result, this system offers to meet specifications to highlight deficiencies and improve performance in areas of the production of items (weapons, military uniforms, medical supplies and more) and goods (chemicals, fuels and other consumables), which directly addresses the DLA procurement requirements to provide accurate evaluations of performance in value-added, non-valued added and total energy computation along with its associated emissions. In Phase I, RedShred will provide a proof of concept to demonstrate the feasibility of LLM modeling of manufacturing processes by delivering an information technology system capable of calculating the value-add, non-value-add and total energy in a scalable manner, for a minimum of 15 distinct manufacturing procedures, reaching TRL 3. In Phase II, RedShred plans to build a working prototype capable of confirming the estimates and providing preliminary cost and pricing data to fulfill selected DoD and commercial business cases.  This prototype will demonstrate the maturity of the RedShred platform alongside TRACEE to address the problem of energy calculation to guide optimizations in the procurement process.