AI Engineering
Practical AI, engineered for real workflows
We treat AI as a component to design, test and monitor, not a demo. Where it saves time or improves decisions, we build it in. Where a simpler solution works better, we say so.
What we build
AI capabilities for your business
Services we design and build for clients. Where a capability already runs in one of our own products, it says so.
-
Service we build
AI workflow integration
Language models connected to the tools your team already uses, at the step where they save the most time.
-
Service we build
Document processing and structured extraction
PDFs, scans, forms and dictation turned into structured, validated data your systems can use.
-
In production: Radiora Medical
AI-assisted reporting
Draft reports structured from notes or dictation, for a qualified person to review and approve.
-
Service we build
Bilingual knowledge search
Answers found in internal documents in French and English, with the sources shown.
A grounded example
Radiora Medical: from dictation to a signed report
In Radiora Medical, a product we built, radiologists dictate from a computer, a phone or an audio file. The AI structures the dictation into a draft with Indication, Technique, Results and Conclusion.
- The AI never diagnoses and never signs.
- The radiologist reviews, edits, validates and signs every report.
- Data is isolated per establishment, with a complete audit trail.
Dictation
Abdominal CT without contrast for right upper quadrant pain. Liver normal in size. No biliary dilatation. Simple two-centimetre cyst in the right kidney. Conclusion: simple renal cyst, no other significant finding.
Structured report, draft
- Indication
- Right upper quadrant pain.
- Technique
- Abdominal CT without contrast.
- Results
- Liver normal in size. No biliary dilatation. Simple 2 cm cyst, right kidney.
- Conclusion
- Simple right renal cyst. No other significant finding.
How we keep it reliable
Safeguards designed in from the start
-
In production: Radiora Medical
Human-in-the-loop approval
High-stakes outputs stay drafts until a qualified person reviews, edits and approves them.
-
How we work
Evaluation and quality assurance
Test sets built from real cases; accuracy, tone and failure modes checked before launch and monitored after it.
-
In production: Radiora Medical and EduSen
Privacy, access control and auditability
Access control, isolation between organisations and audit trails, so every record and action can be traced.
-
How we work
Operating-cost management
Right-sized models per task, caching and usage limits, so the cost per request is known before you scale.
Where to start
One workflow, one number, then scale
Pick a workflow and a metric
One process and the number it should move: time per task, error rate or response time.
Prove it on real cases
A prototype tested against a set of real, anonymised examples before anyone relies on it.
Roll out with oversight
Human approval where the stakes are high, monitoring, and a cost per request you can plan for.
Have a process that takes hours every week?
Describe it in a few sentences. We will tell you honestly whether AI is the right tool.