After exhausting prompt-engineering improvements, Meridian's data science team still finds the general-purpose base LLM struggles with the airline's specific maintenance terminology and internal procedure codes. What is the appropriate next step, and what should the PM understand about it?

Fine-tuning — further training the base LLM on Meridian's own curated maintenance-terminology and procedure-code data — is the appropriate next step once prompt engineering alone can't close the domain-specific gap, and the PM should understand it requires more investment (curated data, compute, review) than prompt engineering but produces a model better specialized to Meridian's vocabulary. Claiming fine-tuning is never appropriate for an airline is wrong and contradicts the exam content outline's explicit inclusion of "implement fine-tuning of LLMs for specialized domains" as an in-scope PM competency. Permanently disabling the assistant instead of fine-tuning it is an overreaction that discards the earlier progress and value already demonstrated by the tool. Claiming fine-tuning removes the need for human review is wrong and dangerous — fine-tuning improves domain fit, it does not eliminate hallucination risk or the need for verification in a safety- and compliance-sensitive context like dispatch procedures.