Laureate of the Convergence IA Challenge, France 2030

MarIAnne is a sovereign AI in production that turns an institution's knowledge into queryable intelligence, under human supervision. Built for organisations where precision, compliance and traceability are non-negotiable.

answers correct on substance, internal expert evaluation across 55 cases

positive feedback, test session of 9 June 2026 with the CNCAC teams

improvement in accuracy between versions (2.72 → 4.79 / 5)

indexed chunks, 7 source types

Pass on expertise, guarantee fair answers

The 2015 reform doubled the notarial workforce in Normandy: 194 offices, a third created after 2015, 415 notaries with 255 appointed after 2015. Hence an urgent need to pass on the tacit knowledge of experienced practitioners and to guarantee fair answers on sensitive ethical questions.

GenieFactory deployed MarIAnne with the CNCAC: a system that captures notarial knowledge and returns it in real time, citing the exact source. It flags when a question falls outside its scope rather than making things up.

Live session of 9 June 2026: 58 conversations, 148 answers generated, 51 evaluated, 80% positive feedback. Across 55 internally evaluated cases, roughly 9 in 10 answers are correct on substance. The one improvement identified (displayed sources) had already been fixed on 7 June.

Not a demo. In production.

The Convergence IA Challenge is a public innovation programme run under France 2030. Season 2 asked a fundamental question: how do highly regulated institutions actually adopt AI, not in a demo, but in production?

GenieFactory and the CNCAC answered with MarIAnne, an already functional system. The challenge was labelled among the selected challenges of Season 2, recognising its maturity and its potential for national replication (CSN, more than 80 chambers) and international replication (ANF network, 30 member states).

Sovereignty

Hosted in France. Client ownership of code and data. No vendor lock-in.

Compliance

Built in from the design stage, not bolted on at the end. Full logging of exchanges. Human supervision at every correction.

Continuous improvement

Every negative feedback triggers an expert review, a supervised correction and an update to the knowledge base. The knowledge lives and improves.

The Workbench

Documentary capitalisation

Business teams upload, organise and validate sources. Experts create reference sheets that hold authority over the answers.

The Forge

Knowledge engine

Hybrid RAG backed by a Neo4j graph. Semantic chunks enriched with relations (in-force / repealed text, national / local rule), indexed. Every answer cites the exact source.

Continuous evaluation

Improvement loop

A negative feedback triggers a review by the business expert. The error is classified, a correction is drafted and integrated under supervision. The system gains precision without retraining a model.