AI Development
We develop enterprise AI systems and computer vision systems — from reading your regulations and process to running in production. Commercial, open-weights or custom models, inside your perimeter when the data requires it, with quality measured before launch.
Discuss your use caseWe build AI systems for a specific customer process: language models connected to enterprise data and tools, and computer vision systems that turn video into verifiable events. We own the whole chain — data, models, rules, interface, quality and operations in your environment.
RAG over knowledge bases, AI agents for business processes, staff assistants, document extraction — on commercial, open-weights or custom models, inside your perimeter when needed.
Detection, classification, segmentation and tracking, with a state machine and an event model on top of your rules. Video analytics, rule enforcement, image-based measurement.
Our own product: a governed AI agent for identity administrators, vendor-neutral over midPoint, SailPoint and other IGA platforms.
We start with your regulations, documents and how the process actually runs. Rules and edge cases are formalised before any model is chosen.
Test sets and acceptance criteria are agreed with you after discovery. Every new version is compared with the previous one on the same data.
Models run inside your perimeter when data must not leave it. Data flows, logging and human oversight are designed so your compliance team can assess them against the GDPR and the EU AI Act.
Where an error is costly, the system prepares the evidence and proposes a conclusion; the decision stays with your expert, and their review flows back into training.
No customer or project names — only the class of problem and the approach.
The customer's rulebook formalised into a state machine and an event model; the system finds incidents and prepares a card with synchronised replays for the expert. Stage: pilot development on historical recordings.
Our own health-tech project: video of a blood drop from a device of our own design. About 95% agreement with manual counting on a validation set of 121 frames, roughly 30,000 cells.
Our governed AI layer for IGA: 11 read-only tools over the MCP standard, RAG over access policies, runs with the user's permissions.
Process and rules, data, hosting and security constraints, acceptance criteria.
A working system on real data, checked against the agreed criteria.
Deployment in your environment, quality monitoring, retraining and evolution.
Two kinds of systems. Enterprise AI: RAG over knowledge bases, AI agents for business processes, assistants and document extraction, on commercial, open-weights or custom models, including private deployment. And computer vision: detection, classification, segmentation, tracking and an event model on top of the customer's rules.
With discovery: the process and its rules, the data, hosting and security constraints, and acceptance criteria. Then a pilot on your data, and production only after acceptance.
Yes. We deploy open-weights or custom models on your hardware or in an isolated segment. Commercial APIs are used only where your data policy allows it.
We are identity and security engineers. We build AI systems with access control, action logs, secret masking and measured quality from day one — as part of your enterprise environment, not as a standalone chat.
After the 2026 AI Omnibus, from 2 December 2027 for Annex III use cases and from 2 August 2028 for Annex I products. Obligations for general-purpose AI model providers have applied since 2 August 2025. Whether your system is high-risk depends on the use case and is assessed by your compliance team.
