We build financial decision infrastructurefor people who decide without an analyst desk behind them.
Cross-referenced, dated data, calculation that can be redone by hand, opportunity cost always in the account. No recommendation.
There is a layer that separates deciding well from deciding in the dark.
Cross-referenced, dated data, alternatives compared on the same scale, explicit priority, opportunity cost calculated.
In practice the obstacle is always the same: the information exists, but it arrives in documents of inconsistent format, without a reliable date and without a common scale. Someone has to read, structure, check and compare; and outside the institutions, that someone does not exist.
Inside large institutions this layer has existed for decades.
Outside them, for individuals, independent investors, developers, chains and operations of any size, almost never.
Four engineering rules that hold for everything we build.
The number never comes from a language model
Every score, projection and ordering comes from a deterministic rule, written and versioned. The same input produces the same output, always, and any calculation can be redone by hand.
We use language models to read documents, explain results and review the deterministic decision; never to calculate. When the model diverges from the rule, the divergence is recorded and audited.
Every data point carries source, date and confidence level
No number appears alone. Each one comes with its origin, its collection date and the confidence of that information in that particular cut.
Low confidence modulates the presentation, not the result. Weak data becomes a result with a warning, never a bad result; otherwise the tool would punish precisely those with the least public data coverage.
Uncertainty is marked, not hidden
When a source is out of date, secondary or diverges from another, that appears marked in the material, next to the number. We do not estimate a value without documentary basis.
The same discipline applies to the market studies we produce internally before deciding to build anything. A document that does not distinguish what is known from what is assumed is no use for deciding.
Opportunity cost is mandatory
Return without comparison means nothing. Every projection is presented against the relevant alternative, with real costs built in, and always as a range; never as a single number.
None of this is a marketing differentiator. It is the minimum an analyst desk would do, and it is exactly what does not exist outside the institutions.
What the engine delivers.
Long-term rental
78/100
Data confidence: highWhat weighed most- Demand+18
- Economic fabric+11
- Infrastructure−6
Residential subdivision
41/100
Data confidence: mediumWhat weighed most- Legal risk−21
- Infrastructure−9
- Trend+7
The same region receives opposite scores depending on the declared intention.
Hover the chart to see the values for each year.
Every return is presented against opportunity cost.
| Debt | Balance | Monthly interest | Reason for the position |
|---|---|---|---|
| Rent in arrears | 4.200 | — | housing |
| Car loan | 18.400 | 1,9% | repossession |
| Credit card | 6.900 | 12,4% | highest interest |
| Personal loan | 11.300 | 3,1% | smallest remaining balance |
Risk comes before mathematics. The highest-interest debt is not first in the queue.
We work on both sides of the balance sheet.
Capital allocation
Intelligence for capital allocation: investment decisions that depend on territory, intended use and horizon.
What enters the calculation- Territory
- Intended use
- Horizon
Financial health and liability management
Organising, prioritising and executing the way out of debt.
How the plan runs- Organise
- Prioritise
- Execute
These are the two markets in operation. The criterion that brought us here is the one we apply to the next ones.
The criterion before the decision.
Before building anything, we produce an opportunity map of the sector: market structure, regulatory obligations with dates, pains verified at primary source, competition mapped, product concepts compared, and a validation plan with objective go or stop criteria.
Every number in those documents carries source and date. What is uncertain is marked as uncertain. Three sectors are at this stage.
Electricity sector
Construction
Agribusiness
The three studies pointed, independently, to the same starting point. It is still early to detail what comes from it; we prefer to talk once it stands.
That is why the choice of the next market is not a bet. It is the same competence applied where the pain is greatest.
Three rules that are not negotiable.
Deterministic, auditable calculation.
No number comes from a language model. Every calculation can be redone by hand, and every data point carries source, date and confidence level.
We explain, we do not recommend.
No screen says buy, sell or sign. The decision always belongs to whoever decides; our job is to make it legible.
Revenue from subscription, never from transaction.
We do not sell credit, we do not sell advertising, we take no commission. There is no incentive to push anything.
The third principle sustains the other two. A product that earns on the transaction always ends up optimising for the transaction.
What we are not.
We are not an investment advisory.
We issue no buy or sell recommendation, and no valuation report.
We are not a financial institution.
We do not originate credit, we do not intermediate transactions, we are not a banking correspondent.
We are not a listings portal.
We show whether a place holds up, not what is for sale in it.
We are not an account aggregator.
We do not ask for bank credentials and we do not access anyone’s account.
We do not replace the licensed professional.
Where the law requires a qualified professional, our product supports that professional; it never takes their place.
We do not sell user data.
Never, to anyone, under any circumstance.
The engine lives isolated from the interface.
The calculation engine lives isolated from the interface, with no framework dependency and its own types and schemas. It is code that can be tested on its own, ported between platforms and verified line by line.
Determinism first.
Business rules in a shared, framework-free package, with test coverage over boundary cases. The interface consumes the engine; it never reimplements the rule.
Auditable extraction.
Every field read from a document carries a confidence indicator and a trail back to the original passage. A doubtful field goes to human review. A document marked valid by mistake is a liability, not a bug.
AI as a language layer, not a calculation layer.
Failover across multiple providers with circuit breaker and deterministic fallback. If they all go down, the product keeps working; it only loses the narrative.
Offline where the field demands it.
Part of the markets we study operate without stable signal. Where that holds, synchronisation with conflict resolution is an architecture requirement, not a feature.
Quota and limits decided on the server.
Nothing that depends on trusting the client.
In operation today.
- ZiageoLocation intelligence for capital decisions that depend on a place.
- PlanoQuitaDiagnosis, prioritised plan and week-by-week tracking of the way out of debt.
Individuals · independent investors · developers · chains · operations of any size
We build for Brazil, on the regulatory and cost reality here.
To talk about the thesis.
The investors page explains the thesis, the method and the real state of the product. The numbers go in the material, after contact.