A voice agent for dispatch
An agent that phones the on-call technician, describes the fault, and confirms whether they can attend and when. It says it is automated, dials only the on-call list, and doesn't record calls.
The outcome
When a site needs a visit, the agent makes the first call: it introduces itself as an automated caller, describes the fault, asks whether the technician can attend and when, and reads each answer back to confirm it. Verified end to end in live test calls.
The situation
The hourly report already knew which sites needed a technician and why. Getting someone to the site still meant a person phoning down a list, explaining the fault from scratch and writing down a time.
What we built
A SIP caller on a server with a public address, so call audio works in both directions. The technician's speech is transcribed, the reply is spoken by a text-to-speech model, and a language model runs the conversation from a short script: who is calling, what the fault is, can you attend, when. The agent waits for the person to answer before it speaks, lets them interrupt, and repeats each answer back.
The limits are part of the design. The agent can't change anything on the network; it only reports. It dials only numbers on the on-call list, identifies itself as automated on every call, and doesn't record calls, which keeps it straightforward under POPIA and RICA. The first call after any change goes to a test number.
What changed
Dispatch can start with a call that already knows the fault, instead of a person reading it out from a report.
Need something like this?
Tell us what you’re working on. You’ll get a straight answer on whether we’re the right firm for it.
Start a conversation →