RAG for Aircraft Manual Comparison: Compliance and Version Control

The major US airline is the world's third-largest airline, operating one of aviation's most asset-intensive, safety-sensitive, real-time networks. The a large application estate is anchored by a large microservice substrate, with thousands of apps involved in dispatching a single aircraft and hundreds of unique services triggered by one boarding pass. Loyalty (their loyalty program with tens of millions of active members), traveler commerce (their advertising platform, hundreds of millions of unique travelers, millions of daily website visitors), and a multi-year applied-AI program sit on top of the core airfare business. Taller's engagement at this airline has been continuous since March 2023, sourced through a staffing partner. The dominant footprint spans four AI / data workstreams that Volume II captures plus two .NET / Angular modernization workstreams. Multi-year SOW extensions have been confirmed across cohorts. Taller engineers placed at this airline have earned strong client-side recognition, with client stakeholders reporting strong satisfaction with team performance.

Aircraft maintenance manuals at a major airline run into hundreds of PDF and XML documents, each updated by manufacturers on irregular cadences. Identifying critical changes between versions had been a manual page-by-page review process, taking weeks of skilled time and a non-trivial source of regulatory exposure when changes were missed. Compliance risk and operational cost are both bound by the manual nature of the comparison process.

Taller implemented the retrieval-augmented system that compresses version comparison into minutes of automated analysis with full traceability of every detected change. The architecture decomposes each manual into the semantic chunks compliance reviewers actually care about — procedures, parts numbers, torque specifications, safety callouts — and compares versions at the chunk level. LangGraph orchestrates the comparison workflow as a directed graph of retrieval and re-ranking steps; knowledge graphs hold the structured relationships between procedures, parts, and aircraft variants so that a change in one document propagates to the right downstream reviews; AWS Bedrock supplies the LLM substrate with the data-residency and regional-availability properties an airline's compliance function requires. The output is a structured changeset that names the affected procedures, the prior text, the new text, and the regulatory category each change belongs to.

The deployment reduced manual document processing time from weeks to minutes, improved accuracy in identifying critical version changes, and delivered full traceability of changes for regulatory compliance.

Compliance is the regulatory perimeter of an airline's operations, and any AI deployment that reduces compliance risk while reducing operational cost is a structural improvement in the airline's safety-and-economics posture. The aircraft-manual RAG is the flagship reference for production AI in highly-regulated, document-heavy operations across transportation and adjacent industries.