RAG for Aircraft Manual Comparison: Compliance and Version Control

A major US airline engaged Taller through a staffing partner in March 2023, with work spanning applied AI and core modernization. Taller ran six pods across AI/data and .NET/Angular rebuilds, with multi-year extensions confirmed.

The airline’s aircraft-maintenance manuals comprised hundreds of PDF and XML documents, each updated by manufacturers on irregular schedules. Finding the critical changes between versions had been a manual, page-by-page process: weeks of skilled time, and a real source of regulatory risk whenever a change was missed. Both the compliance risk and the cost were tied to the manual nature of the process.

Taller built a retrieval-augmented system (RAG, an approach that grounds AI answers in the source documents themselves) that compressed version comparison into minutes of automated analysis, with a full trace of every change it detected. The system broke each manual into the meaningful pieces compliance reviewers actually cared about (procedures, part numbers, torque specifications, safety callouts) and compared versions piece by piece. LangGraph ran the comparison as a series of retrieval and ranking steps; a knowledge graph (a structured map of how procedures, parts, and aircraft variants relate) ensured that a change in one document flagged the right downstream reviews; and AWS Bedrock supplied the underlying AI models with the data-residency and regional-availability an airline’s compliance function required. The output was a structured summary naming each affected procedure, the old text, the new text, and the regulatory category the change fell under.