Service · published 23 Aug 2026
AI that survives an audit.
Our production track record is aviation — one of Europe's most regulated domains: retrieval over controlled manuals with page-level citations, nine-department review workflows with full audit trails, and regulatory change-monitoring that runs daily.
AI for regulated industries is system design where traceability is a functional requirement: every answer cites its source, every automated decision has a human review path, every change is logged, and the regulator's question 'why did the system say this?' always has an answer.
Traceability as architecture
In our aviation systems, every answer carries an inline citation to the exact source page, rendered beside the controlled document itself; operational claims trace to the system of record. Detected regulatory changes become tracked work items with status, assignee, and a complete review history. This isn't compliance theater bolted on — it's the retrieval and workflow architecture itself.
The regulatory-change problem, solved twice
We built regulatory change-monitoring first as a production module for an airline — four scraper architectures, content-hash change detection, AI summaries of only what changed, routed through nine departments — then productized it as RegNotify, which today tracks 5,000+ EASA regulations across 2,764 aircraft types, ingesting every six hours. Few teams have shipped this pattern once; we operate it continuously.
Questions we actually get
- Can you use LLMs in a regulated environment at all?
- Yes, with the right system boundaries: retrieval-grounded answers with verifiable citations rather than open generation, human review on consequential outputs, logged interactions, and evaluation sets that document system behavior. The model is a component inside an auditable system — the system is what gets regulated.
- How do you prevent an AI system from giving unsafe answers from official documents?
- Grounding plus refusal: answers must be supported by retrieved content from the controlled document set, a sufficiency check refuses rather than guesses when support is weak, and citations let the professional verify the source page in one click. In aviation, our users are trained staff verifying against manuals — the system accelerates them; it doesn't replace their authority.
- What about MDR, EASA, or other certification regimes — can you handle those?
- Honestly: we build compliance-ready software architecture — audit spines, traceability, documented behavior — and we've done it in EASA-regulated aviation operations. For formal certification programs (like medical MDR), we work alongside specialized regulatory partners rather than claiming certification expertise we don't hold. That division of labor is in your interest.
- How do you keep up with regulatory change automatically?
- Scrape the official sources on schedule, detect changes by content hash, resolve linked documents, and have an LLM summarize only the changed sections into what changed, why it matters, and what to do — then route it to the people who must act. Our production cost for this pattern is about $0.02 per weekly run per source set; RegNotify runs it every six hours at corpus scale.
- Which regulated domains do you know?
- Deep, shipped experience: European aviation (EASA, German LBA, Part-145/CAMO operations). Adjacent, built-but-lighter: EU food-labeling compliance. For other regimes, the architecture patterns transfer; the domain expertise we build with your specialists — and we'll tell you plainly where our knowledge ends.
- What does an audit trail for an AI system look like?
- Every interaction logged with its retrieved sources, model version, and output; every automated detection linked to the human review that followed; corpus changes versioned and deliberate. In our systems this exists as admin-facing tracing — per-query confidence, latency, cost, and sources — because operators need it daily, not just when the auditor visits.
The fastest way to find out if this fits: a free audit on one workflow — an honest answer on whether it's worth building, and a written summary you keep either way.