Precursors,before they're patterns.
Specialist agents read your operation as it moves: the classifier picks the ICAO category, risk drafts the score, the precursor surfaces the cluster building underneath, and eAvy answers your questions in plain English. Every run cites the records it drew on, a person validates it, and you can replay it years later.
Read the operation
like a weather map.
Risk Weather is the one view that tells you where pressure is building before it becomes an incident. Each area of the operation reads calm, worth a look, or building — drawn from the signals across every module, not a single tab.
Cited at run.
Cited at audit.
Every agent run captures its inputs, output, confidence and the source records it cited on a locked row. Years later, the auditor reads exactly what the safety officer saw — not a regenerated approximation.
It knows the difference
between a blip and a trend.
Your safety indicators are watched on a control chart — the same statistical method used on a flight deck and a factory floor. A point inside the normal band is noise; a point that breaks the upper limit is a real change, and that is what raises a flag. No chasing every month-to-month wobble.
Five moments.
No black-box appeals.
- When an occurrence is classified.The classification is captured the moment it runs.
The classifier reads the narrative against your taxonomy and writes one answer — with what it looked at and which records it referenced attached. It never silently re-runs. If the read changes, that is a deliberate, signed event with a reason.
- When you ask eAvy a question about the operation.A cited answer, with the underlying records attached.
Ask in plain English. eAvy answers with the records it drew the answer from — not “according to recent data” prose. Every assertion is reviewable, and every citation opens the source occurrence, finding or audit.
- When the regulator asks how the AI made a call.Open the run. The full trail is right there.
No black-box appeals. Every AI run is a record: what the AI saw, what it concluded, its confidence, the cited source records, the time, and the person who validated it. Open the run, and the auditor sees exactly what the safety officer saw on the day.
- When a pattern is building across the operation.The precursor surfaces what you have not asked yet.
Latent conditions lining up across reports, hazards, audits and barriers — drawn together across modules, not stranded one report at a time. The precursor raises the cluster as a typed finding, so the post-holder gets a record, not an inkling.
- When the AI is wrong.An override is a signed event, with the reason captured.
AI gets things wrong. When it does, the override is a record — who overrode it, when, why, and against what new evidence. The trail is part of the workflow, not bolted on as an export afterwards.
One copilot
across the whole picture.
eAvy sits on top of every agent. It reads a record, traces what it’s linked to, finds the look-alike occurrences and drafts the next step — always as a card for you to accept, edit or reject. The copilot proposes; a person decides.
Agents don't live
in a separate dashboard.
Every analysis attaches to the record it analysed, and every alert routes to the post-holder who owns it. Four examples that matter the day a regulator asks for the trail.
See the intelligence layer.
On your data shape.
Walk a classification run, an override, the Risk Weather view and the audit replay — with cited sources and a person in the loop on every one — in 30 minutes.