A Specialized Language Model. Everyone reads SLM as Small. The only small thing about Aless is what it costs to run and to keep running. It is not small in what it knows. The S that matters is Specialized.

What specialises it is people. A field is only entered once practitioners are at the table: a minimum of twenty-five years in that field, paid properly for their time. They set what the model has to learn, and they judge whether it has learned it.

We ask how they really decide — the real cases, the nuances, the exception that shows up twice a decade. That judgement goes into the learning itself.

Banking came first. It was designed by people who spent their working lives inside large banks, across several countries.

That is the Human Touch, and no field is entered any other way.

The AI reads your question and finds the right route. It does not do the sums. Tested, checkable code does that — so a number is computed, never guessed at.

No figure exists without a record behind it. Open any number and you get the workings: the source record, the rule applied, the table it came from, the log entry that wrote it. A figure that cannot be traced is refused rather than invented.

And the work runs overnight, so the answer is usually ready before you ask. Ask something genuinely new and it is built for you, kept as yours alone, and run every night from then on.

In the banking demonstration today: 57 pre-fabricated modules, an encyclopedia of 7,211 entries that routes any term to its module, the optimizer and reverse solver, the recommendation board, ready-made dashboards and a composer for your own, a chat box with Bambi, and the night batch’s own receipt. What is built today →

The five levels, in full →

Which fields, today? Four: Financial Institution — AFIAS, the Aless Financial Institutional Analytical System for senior management (banks first; asset managers, pension funds, hedge funds), Education, Medicine and Architecture. The same Aless in each, set up for that work.

Built into all of them: data scrubbing — your data in whatever shape you have it, including legacy, out in one clean form; data synthesising — a whole institution built to test against before anything real is touched; and encryption throughout.

All four →

Where? Inside your own building, on hardware you have. It does not run in somebody else’s datacentre and it does not need one.

You take the depth you need — the whole enterprise, a division, a department, a group, or a single book. It is modular all the way down, and a module carries the language of the place it is deployed in.

And where it could never be bought, it is given. Once the commercial side is developed, the same foundation will be used to build a separate philanthropic corporation, operated with funding from its own sources — donations, government grants and the like — never from the investment in the commercial business. Aless · The Human Touch →

Because everything above only holds if nobody outside can reach it. An AI reasoning over your whole institution is the most concentrated target you own. So the safety is not a feature of the AI — it is the system the AI sits inside.

How that system works. No wifi. No outside ethernet. A physical data diode sits at the boundary: your data travels in, the answer travels out, and nothing travels back the other way. That is how the hardware is built, not a setting anyone can change — so there is no way back in for anyone to use.

Everything is AES-256 encrypted from the moment it is cleaned and again before it is stored. Aless works on tokens rather than your real figures, so it could not disclose one even if it were told to.

And the keys are not in the machine. They live in hardware security modules — YubiHSM 2, FIPS 140-3 — two per protected machine. A key never exists as a file, so it cannot be copied off a disk.

Which means the hardware is worthless on its own. Take the servers out of the building and what you have is encrypted noise: without the keys nothing decrypts, nothing runs, and nothing can be read. A stolen Aless is a brick.

Specified in the design; the modules are not yet purchased. We would rather say that than badge hardware nobody owns.

Every action is written to a log at the rack that cannot be edited. Reading it means standing in the room. And what leaves is pure data — text and numbers, never code.

The whole system, drawn →

Who teaches Aless? The people who spent their lives in the field. Before it enters banking, or any other field, we sit down with recently retired professionals — people with twenty-five years or more behind them — and ask how they really decided: the real cases, the judgement calls, the exception that shows up twice a decade. That knowledge is transferred into the system, not typed in from a manual. That is the Human Touch.

Two halves, working together. The human knowledge is the qualitative half: what matters, what to watch, what an experienced person would do. The Structural Equation Module is the quantitative half: it holds the whole institution as one connected system, so when anything moves — a rate, a deposit, a loan book, a rule — it computes how every other measure responds. Behind it sit the advanced calculators we built for exactly this: solvers that find what has to move to reach a target, optimisers that pick the best mix within the rules, stress engines that shock the whole system at once and show the path back, and probability and sensitivity methods that say how sure an answer is.

Why it feels different. Ask about one thing and you get the whole picture: the accounting behind the number, the banking rule that governs it, the tax and regulatory consequences, and what else in the institution moves with it. The numbers come from the calculators; the meaning comes from the people. Together they give a recommendation a senior manager can act on, with the reasons on the face of it.

Built in advance. Nine-tenths of what an institution needs is ready before it arrives, as pre-fabricated modules specified by the professionals who ran those functions; the institution adds its own tenth. Finance comes first, then education, medicine, architecture and law.

The Human Touch is how Aless enters every field, and no field is entered any other way.