OKDeep learning where classic vision falls short
Seals, surfaces, natural or highly variable products: defects that cannot be described with rules. Neural networks trained on your own products, integrated into the line.
Home / Artificial intelligence
Artificial intelligence · bespoke development
Artificial intelligence that works: on the line, in maintenance and in the office
We don't sell AI as a promise. We develop it and use it every day: in vision, in machine maintenance, in the analysis of production data and in our own sales management. What has delivered results in our own company, we bring to yours, tailored to you.
How we see it
Concrete tools, not slide decks
Useful artificial intelligence looks very little like the one in the headlines. It is a model that tells a good seal from a bad one, a warning that arrives before a motor stops, a report that is generated automatically or an assistant that answers based on the machine manual. That is what we do.
- We start with the problem, not the technology
- We work with your data and your machines, without changing what already works
- Every project is measured: fewer stoppages, fewer rejects, fewer wasted hours
- We develop it ourselves, with close support in your language
- You decide how private your data stays
AI under control
AI agents that are measured, have limits and answer to a person
TIVOX runs every day on more than twenty AI agents across five departments. Every run is measured, every agent has limits on steps, time and spend, whatever a person corrects becomes a rule, and nothing reaches a customer until someone approves it. This is the method we bring to your company.
TIVOX IALiveAI agents · the team at work
AI agents · the team at work
- OK
- Working
- Awaiting approval
- Limit reached
- Idle now
Mock-up with demo data of the system we use at TIVOX.
- MeasuredEvery run, on record
- Duration, cost, steps and outcome of every agent, every day. If it fails or stops halfway, it is flagged in red and the person responsible is alerted.
- LimitsLimits that cannot be exceeded
- Each agent has a maximum number of steps, time and spend. When it reaches it, it stops and reports it: the monthly spend never goes past the limit.
- LearnsEvery correction, a rule
- What a person changes or rejects, with the reason, becomes a rule the agent reads before it works again. Past mistakes are checked on every draft.
- ApprovalA person decides
- Anything for a customer or the public only goes out with “Approve”. “Reject” and “Edit” are one click away, by email, on screen or from a phone.
Where we apply it
Six fields where it is already delivering results
These are examples of what we have running today. Each one is adapted to your plant, your product and your systems.
Sensors that warn before the breakdown
Accelerometers, temperature and humidity probes, power consumption. Once there is data, the AI learns the normal behaviour of each motor, gearbox or bearing and warns of any abnormal behaviour, and not only in motors: any equipment that generates data can be monitored.
An assistant that knows the manuals
It consults the manuals, catalogues and diagrams of your machines and drives and guides the diagnosis step by step. Less time spent searching through documentation.
An analyst that reads your data
It summarises the week, finds patterns in stoppages and waste and points out where to act first. Automatic reports, per shift, day or week, in natural language.
Bots and agents for sales management
Quote follow-ups that are never forgotten, prospecting with agents that prepare the information before every call, emails and reminders drafted and scheduled automatically. We use them ourselves every day.
Automatically generated reports and documents, with your approval
Production reports, data sheets, procedures or quotations: an agent drafts them from your data and documentation, and a person reviews and approves them before they go out, in whatever language is needed.
Predictive maintenance
A breakdown announces itself. AI listens.
A motor running hot, a bearing starting to vibrate, power consumption creeping up. With accelerometers, temperature and humidity probes and electrical consumption readings, artificial intelligence learns how each piece of equipment normally behaves and warns you when something changes, before the line stops.
- Temperature, vibration, consumption sensors and more
- AI learns the normal behaviour of every motor, gearbox and bearing
- Instant alerts to maintenance, with the reason
- MTTR and MTBF measured, not estimated
Why did line 2's OEE drop on Tuesday?
OEE by day · line 2 · %
On Tuesday line 2 was stopped for 1 h 52 min: 1 h 10 min for a format changeover that overran and 42 min of micro-stops at the labeller. Without that changeover, OEE would have been 77%.
Source: line 2 stops on Tuesday and the changeover log
Mock-up with demo data.
Data
Any data in your company can start to talk
Production, quality, maintenance, purchasing, sales, energy. Almost every company stores more data than it uses. We connect AI to your databases, spreadsheets and current systems and give you answers in plain language: what happened, why, and what is worth doing about it.
- Questions in plain language, answers from your data
- Periodic reports that are generated and sent automatically
- Alerts when an indicator goes out of range
- Integration with what you already have: ERP, MES, SCADA, spreadsheets
- Models on your premises or in the cloud, as you decide
Our own development
A company that runs on its own AI agents
Before offering it, we use it ourselves. TIVOX runs every day on its own platform with more than ten connected tools and artificial intelligence agents that prepare the work so that a person can review and approve it. Everything you see here can be set up in any company, whether industrial, service or distribution.
Everything in one place
A platform with encrypted access, from the office or from a mobile phone. Each person has their own permissions and each tool has its place.
AI prepares, you decide
Agents work autonomously according to their schedule: every morning, every Monday. Nothing goes out until someone clicks "Approve", "Reject" or "Edit", by email, on screen or from a mobile phone.
AI that learns from your corrections
Every run is logged with its time and cost, with spending caps. Whatever a person corrects becomes a rule the agent reads before it works again.
On your premises and backed up
Your own databases, automatic backups and a history of every change. The AI works on your systems; the data does not leave your infrastructure.
The tools
What they do for us every day
We built them for TIVOX, but none of them is specific to machine vision: they cover sales, customers, marketing and administration, which every company has.
TIVOX HUBLiveLive map · one report per minute from every line
Live map · one report per minute from every line
- Producing
- Attention
- Stopped
Mock-up with demo data: customers, lines and figures are fictitious.
Customers
Every installation, on one map
Each installed vision system reports automatically, every minute: cameras, production of the day, alerts, version and licence. Updates are released through channels, first to a test unit and then to customers. The customer's equipment is not accessed from outside: the connection always goes out from it.
Sales
Prospecting with agents that do the research
It starts from a public register of companies and turns it into a market map. The agents research each company online, find out who makes the decisions using public professional data, find real reasons to call, always with their source, and prepare the day's call list. The salesperson goes into every call knowing who they are talking to.
Alongside it, the CRM: quotations, contacts and activity log, with automatic follow-ups that wait for approval before they go out, and a mobile version.
Marketing
A website that looks after itself, with your permission
Every Monday an agent measures how Google sees us and proposes specific changes, each backed by the figure that justifies it; they are applied with one click. Then, week after week, it checks whether each change has worked.
An assistant in your pocket
A Telegram bot to check what is pending, approve and receive alerts from your phone; and WhatsApp alerts when something needs attention.
Quotations, expenses and tasks
A quotation configurator, expenses from a photo of the receipt, a task list shared between people and agents, and a map of everything running in the company.
Beyond TIVOX
A librarian for all your documentation
We are already taking this method to other companies, for example to a service, spare parts and machinery sales company, for which we are developing a "librarian": an assistant that knows all its documentation (manuals, exploded views, data sheets, procedures and quotations) and answers in seconds, citing the document each answer comes from.
- Finds the part, the manual or the procedure without knowing where it is stored
- Every answer states which document it comes from, so it can be checked
- Each person sees only what they are meant to see
- Learns from the corrections of the people who use it
TIVOX ATLASLive
Five people from quality took the 8-hour workshop “Machine vision for quality” in September. Four passed the assessment; the fifth has the Cpk module pending.
Sources
- 1Outline of the machine vision for quality workshop
- 2Attendance record, 16/09
- 3Workshop assessments
Mock-up with demo data: documents, figures and questions are examples.
How we work
From idea to result in weeks, not years
An AI project does not have to be big or expensive to prove that it works.
What costs you time or money
A visit or a video call to understand the process, the data available and the number you want to improve.
With your real data, in a few weeks
A first, well-defined model with a fixed budget, to see whether it adds value before investing more.
If an indicator does not improve, we do not continue
It is deployed as part of your daily routine, with training for your team, and what the AI learns is reviewed with you.
Your idea
If it can be described, it can very probably be developed
New capabilities appear every month, and what was impossible a year ago can now be solved in a short time. In artificial intelligence the limit is set by the idea, not the technology. If you have a process that AI could do better, we study it with you and, if it adds value, we make it happen.
Frequently asked questions
Answers to the most common questions
Do I need a lot of data to start?
Not always. In vision, a few dozen good and bad parts are enough to start training; in predictive maintenance, a few weeks of sensor data let the AI learn normal behaviour. We tell you honestly whether there is enough data or what is missing.
Does my data leave the company?
That is your decision. We can work with models that run on your premises or with cloud services under contract, and we explain it before we start.
Does AI replace my team?
No. It frees them from repetitive tasks and gives them the information in advance: the technician decides with more data and the salesperson calls better prepared.
How much does it cost and how long does it take?
We start with a well-defined prototype, a matter of weeks, with a fixed budget. If it works, it is scaled up; if not, you will know soon, without unnecessary investment.
Does it work with my current machines and software?
Yes. We connect to what you already have (PLCs, sensors, ERP, spreadsheets) without changing it.
TIVOX Academy
Help your team understand AI before using it
Hands-on workshops for management, quality, maintenance and office: what AI can do in your factory today, what it can't, and where to start. At your plant or remotely, with examples from your own line.
Let's talk
What would you do with an AI that knows your company?
Tell us about the process that consumes most of your time. We will propose a prototype and a way to measure the result.