Mosaic ATM is excited to announce the launch of the Performance-Based Flow Management (PBFM) Benchmarking Large Language Model LLM Evaluation project, PBLE. This project aims to evaluate whether artificial intelligence can facilitate the implementation of the PBFM concept and improve Traffic Flow Management (TFM) decision-making.
The goal of the PBFM LLM Evaluation project is to enhance management of the Air Traffic Management (ATM) system by establishing performance-focused metrics and measures for the National Airspace System (NAS). These performance metrics are intended to capture the overall expectations of airspace users in areas important to them, such as maximizing available capacity, providing consistent and predictable performance, and optimizing flight efficiency. This approach enables airspace users to set performance targets and priorities for these areas.
Once the project is established, FAA Traffic Managers can evaluate tradeoffs between different Traffic Management Initiatives (TMIs), using data-driven solutions to identify approaches that best meet airspace user preferences, improving efficiency.
Figure 1. TMI Tradeoff Analysis
Our objective for the PBLE project is to leverage a Large Language Model (LLM) enhanced with Agentic Retrieval-Augmented Generation (RAG) technology. First, we will construct a comprehensive knowledge graph aligning airspace user service expectations (e.g., Capacity, Efficiency, and Predictability) with historical and real-time FAA data.
Figure 2. Entity and Relationship Extraction
AI agents will then develop software capable of continuously monitoring the NAS, proactively identifying patterns, and suggesting optimal management strategies. Once implemented, this capability will be evaluated for its effectiveness in measuring how Traffic Management Initiatives (TMIs) meet user performance targets.
This LLM evaluation project aligns with the FAA’s strategic objectives of modernizing air traffic management through performance-based frameworks and intelligent automation. Mosaic ATM is dedicated to advancing FAA objectives by combining its extensive knowledge of FAA operations and systems with its expertise in AI and data analytics to address critical aviation challenges.
