Service
Vision and voice systems
Vision and voice systems have to work in your environment rather than in a lab: poor light, dusty lenses, noisy lines, accents. Zero-Shot builds camera and speech systems from data collected on site, and measures accuracy under the worst conditions rather than the best.
What we build
- 01In-line image processing: counting, alignment, surface and label checks
- 02On-site setup including camera placement, lighting and lens selection
- 03Speech recognition and synthesis measured at real telephone line quality
- 04Voice assistants that recover from interruptions and hand over to a human
- 05An edge deployment that runs without a cloud connection
How it runs
01
Data from the floor
Images and audio are collected from the environment before anything is written, deliberately sampling shift changes, lighting changes and bad conditions.
02
Measure the worst case
Accuracy is reported separately for normal and worst conditions. Averages alone produce surprises in production.
03
Pilot on one station
One line, one camera or one phone number goes live first. The false alarm rate is tuned here.
04
Roll out and maintain
Then it spreads to other stations, with a self-check that catches dirty lenses and lighting drift.
What we measure
- Miss rate and false alarm rate together — either one alone is meaningless
- Accuracy under the worst conditions, which matters more than the best
- Decision latency and whether it keeps up with line speed
- Word error rate measured on your own lines
- Handover rate on voice calls, and the reasons behind it
When this is the wrong fit
If lighting cannot be controlled and cameras cannot be repositioned, a vision system stays unreliable — that is physics, not a model problem. On the voice side, we do not recommend voice assistants for emotional or critical conversations such as complaints, cancellations or health matters.
By sector
The same work runs differently in every sector. The detail is here.
Common questions
Can we use our existing cameras?
Often yes, but image quality is measured first. Security camera resolution and angles are usually insufficient for quality inspection, while they are frequently fine for counting and presence checks. We settle that distinction in the first week.
Does it work without internet?
Yes. Vision models can run on a device at the line, so images never leave the plant. Locally running speech models exist too, though quality drops somewhat against cloud models — and that gap is measured and shown.
Does the voice assistant sound human?
The conversation is natural, but the assistant does not present itself as a person. It states that it is an AI at the start of the call. We do this because it is the right behaviour and because hidden setups damage customer trust.
Other services
What we've written on this
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.