What we do
Six practices, one team.
We don't have an off-the-shelf catalogue. Each practice below is a kind of problem we have already solved in production. You bring the pain point; we tell you which practice it falls under, what it costs and how we'll measure it.
WhatsApp and support agents
When it makes sense
Your team answers the same questions all day, orders arrive as voice notes and loose messages, and customers wait hours for something that should take seconds.
What we deliver
- An agent that answers from your company's documents and catalogue
- Booking, ordering and payment inside the conversation
- Voice notes transcribed and understood
- Hand-off to a person when the case calls for it, with the full history
- A log of every conversation and every AI answer
How we measure
Conversations resolved without a human, response time and orders closed through the channel.
In practice
Jundo: an AI support desk that answers on its own and hands off to a person when needed.
See the projectProcess automation
When it makes sense
Someone spends hours copying data between spreadsheets, email and systems. The process works, but it depends on a person remembering to do it.
What we deliver
- A map of the process as it runs today, and who runs it
- Agents that carry out the repetitive steps
- Approval points wherever a person needs to decide
- Integration with the systems you already use
- Alerts when something falls outside the norm
How we measure
Hours given back to the team each month and error rate before and after.
In practice
Sentinela: nine agents check occupational health records and chase missing items with deadlines.
See the projectDocument reading
When it makes sense
Important information arrives as a PDF, a photo or a contract, and someone has to type it all again before it can be used.
What we deliver
- Field extraction from PDFs, images, invoices and contracts
- Each value shown next to the passage it came from
- A quick human check before saving
- Personal data pseudonymised before it reaches the AI
How we measure
Fields extracted correctly and review time per document.
In practice
Estação Clínica: PDF lab results read by AI and checked by the doctor, value by value.
See the projectPredictive models
When it makes sense
One store runs out while another is overstocked. Inventory, purchasing or pricing are decided by gut feeling and last year's spreadsheet.
What we deliver
- Demand forecasts by product, store or region
- Restocking and purchasing suggestions
- Price and scenario simulation
- A model trained on your company's history, not a generic one
How we measure
Forecast error against actuals, stock-outs and capital tied up in inventory.
In practice
Home-improvement chain with 11 stores: −22% stock-outs and −15% tied-up capital, with a 5-month payback.
Custom systems
When it makes sense
No tool on the market fits the way your operation works, and the team lives on workarounds.
What we deliver
- A system of your own, designed from the operation up
- AI built in where it helps, and only where it helps
- Code, infrastructure and documentation in your name
- A first version in use within weeks, not months
How we measure
End-to-end process time and rework.
In practice
AstroMatCon: a full ERP for building materials stores, with 179 screens and built-in AI.
See the projectData and decisions
When it makes sense
The data exists, but nobody trusts the number, the report arrives late and problems only show up once they've already cost a lot.
What we deliver
- Dashboards measured the way you define
- Alerts that flag risk early
- Plain-language questions about your data
- Processing inside your own server when data can't leave it
How we measure
Time for information to reach decision-makers, and decisions taken from alerts.
In practice
AceleraMatCon: invoices from 61 stores become rankings, ABC analysis and a network P&L.
See the projectWhat we don't do
So there are no surprises.
We don't resell licences
If an off-the-shelf tool solves it, we tell you which one and you buy it directly.
We don't ship showcase prototypes
What we build goes into production and gets measured. A pretty demo that doesn't run doesn't count.
We don't let AI decide alone
Where people or money are at risk, AI proposes and a person approves.
We don't lock you in
Code, infrastructure and documentation are yours from day one.
Not sure which practice your problem falls under?
That's normal. Describe the problem the way it shows up day to day; the triage is on us and free of charge.