Relay
Field work, hands free.
THE PRODUCT
Field work divides attention between the environment, the equipment and information that often lives across several systems. Relay turns voice into a hands-free operational layer, so a technician can recover context, check a procedure and record what they find without stepping away from the job. For each task, it brings together the work order, asset history, approved procedures and outstanding actions. The technician can ask what changed since the previous visit, dictate an observation or prepare a parts request in the same conversation. Important actions still require confirmation, and if the connection drops, pending work remains visible until it can be synchronized.
IN USE
- (01)
The job arrives prepared
Relay brings together the work order, asset history, approved procedures and the technician's permissions.
- (02)
The technician works by voice
They can retrieve a detail, check a step and turn a spoken observation into a note or prepared action.
- (03)
Actions are confirmed and synchronized
Important changes require confirmation, and offline work remains visible until it is sent.
ENGINEERING
Relay builds a job-scoped context package by normalizing the work order, asset history, approved procedures and the technician's permissions. Sources are indexed with version and provenance, and retrieval is constrained to the current asset, job and role to prevent specifications from being mixed. The voice pipeline is adapted to technical vocabulary and noise; when bandwidth drops, prepared context remains available while features that require connectivity degrade.
Spoken observations are transformed into typed action schemas: a note, parts request or status change. Reads can execute directly, but every write remains a draft until explicit confirmation. Each queued action retains an identity and audit state; when connectivity returns, synchronization detects server changes and surfaces conflicts instead of overwriting them. A missing or stale source, or a safety condition, blocks the action. Evaluations measure entity accuracy in noise, source retrieval, tool success, latency, sync recovery and safe refusal.
