Projects

What we have delivered, what we can build.

One project is delivered and demonstrable today: Knowledge AI, which you can see working. The other use cases describe what we can build with the same components, and are never presented as existing products. All follow the same rule: your data stays within your perimeter, and every result can be traced back to its source.

Knowledge Intelligence · delivered and demonstrable

Knowledge AI

A sovereign document search product: it answers natural-language questions about your document base, cites every source down to the passage, stays silent when the source is missing, and runs within your infrastructure, including without an internet connection.

Discover Knowledge AI

Cited answers that a colleague can verify
No answer without a source
Offline operation
Reproducible quality evaluation

Knowledge AI, in screenshots

Screenshots of the Knowledge AI interface, in French, taken on 6 October 2026 on a demonstration instance. The corpus consists of sixteen articles of the French Labour Code (Légifrance, Open Licence), with no client data.

Knowledge AI interface, in French. Question: “What does the company risk if the prevention plan has not been drawn up?” The tool's reply: “Information not found. I did not find this information in your documents. Specify the document, equipment or procedure concerned, for example: according to PROC-MAINT-042.” — No answer without a source
The six passages read deal with the prevention plan and risk assessment; none provides for a penalty when no prevention plan has been drawn up. The tool does not fill that silence: it stays silent. The silence concerns these sixteen articles, not French labour law. The message displayed is still generic: it does not say why the tool is silent.
View the screenshot at full size : No answer without a source

Knowledge AI Documents screen, in French, workspace “Site demonstration — Labour Code”: 3 documents indexed, 0 in progress, 0 in quarantine, 0 failed. The three PDFs of the corpus are indexed, in French. — The demonstration corpus
Three documents indexed: none in quarantine, none failed.
View the screenshot at full size : The demonstration corpus

Knowledge AI audit log screen, in French, read-only: date, account, action, object and result of every sign-in and every administrative action, with the export of answer traceability in CSV and JSON. — The audit log
Read-only: every sign-in and every administrative action is dated and attributed to an account.
View the screenshot at full size : The audit log

What we can build

Five more areas, built on the same components

These use cases are not off-the-shelf products. Each one is designed from your corpora, processes and reference frameworks, in a project that starts with scoping.

Document Intelligence

The problem: Your teams manually re-enter information scattered across heterogeneous documents.
What the product does: The product identifies the type of each document, extracts the relevant fields, structures them according to your data model and links each value to the passage it comes from.
What you get: Structured data, verifiable at the source, integrated into your existing applications.

Data & Analytics

The problem: Management indicators rely on manual consolidations that are late and hard to audit, and every new question about your data waits for an analyst.
What the product does: The product computes indicators according to documented rules, feeds your dashboards and lets you query your data in natural language, showing the data each answer relies on.
What you get: Up-to-date indicators and answers in which every figure can be justified by its source data.

Business automation

The problem: Repetitive, low-value tasks slow down the processing of cases.
What the product does: The product integrates with your existing processes and tools: it prepares cases, proposes a reasoned decision and submits it for validation by a colleague, who stays in control.
What you get: Smoother processing, with no decision taken outside human control.

AI governance and evaluation

The problem: AI systems multiply with no inventory, no measurement of their quality and no control of their risks, while the European AI Act requires them to be documented.
What the product does: The product lists the systems in use, documents their data, purposes and risks, measures the quality of their answers on question sets established in advance and keeps a record of every evaluation.
What you get: An up-to-date map and replayable evaluations that prepare your compliance obligations.

Sovereign AI

The problem: Your data cannot leave your network, and some of your sites have no internet access at all.
What the product does: The product runs on your servers, on-premise, with open models and local indexes; where required, it operates in air-gapped environments, with no connection at all.
What you get: AI uses become possible where online services are ruled out, without any data leaving your perimeter.

One rule for all

Whatever the use case, processing runs within your infrastructure, every result can be traced back to its source and you receive the code. No performance figure is announced before it has been measured on your data.

Your use case is not on the list?

Tell us about your process and your data. After review, we will tell you what we can build, and what we would not recommend building.

Tell us about your project Our method