Every maintenance organisation already owns the corpus an AI assistant needs.
Point 145.A.70 prescribes what the maintenance organisation exposition contains, CAMO.A.300 does the same for the continuing airworthiness management exposition, and the acceptable means of compliance prescribe how each is laid out.
Together with the procedures the exposition incorporates by reference, the applicable maintenance data and the records, that is a controlled body of text that already answers most of the questions staff ask in a typical week.
Lavionic’s new guide, “AI assistant for EASA Part 145 and CAMO organisa0tions”, starts from that observation.
An assistant in this setting is not there to know aviation.
It is there to answer from the organisation’s approved documents and to show the paragraph each answer came from.
When a certifying staff member asks who may release a component after a particular task, the correct answer is the one in the organisation’s own exposition, at the revision the authority accepted.
The guide lists ten questions staff ask in a typical week and, for each, where the controlled answer lives: which clause of Part 145 or Part-CAMO, which chapter of the exposition, which procedure.
It sets out what the assistant may do, which is answer with a citation, draft procedure text, and draft amendments and audit responses for review, and what nominated persons keep, which is the approval, the release to service and the signature.
It also explains why a general chatbot is the wrong tool for this job.
Uploading the exposition into a chat window fixes one conversation for one person.
It does not give the organisation one controlled corpus at one revision, and it does not stop a colleague from uploading a superseded issue.
Point 145.A.45 makes the organisation responsible for holding and using current maintenance data; a tool that cannot show which revision it answered from does not help meet that.
The guide closes with six steps to introduce an assistant: start from the exposition, ingest at the accepted revision, restrict who can change the corpus, define what the assistant may draft and who reviews it, write that into the procedure for AI-assisted work, and measure the answers against the source before anyone relies on them.
The guide is written by Oguz Hicdurmaz, founder of Lavionic and an airworthiness engineer with more than twenty years in manned and unmanned certification.
It is published on wingman360.ai, where Wingman360 Teammate, Lavionic’s AI assistant for aviation compliance and certification, is documented.
