The solution
Patent filed

AI technology

An AI that doesn't guess prices. A language model can understand and write; it should not decide sizing or pricing on its own. DARYL keeps the two apart.

Our proprietary technology

Cognitive Learning

Patent filed

Large language models have one quality nothing else matches: they understand a request phrased any which way, whether a rambling email or a voice note dictated in a noisy workshop, and can turn it into a readable answer. They also have one flaw that is disqualifying in our field: they produce the most plausible answer. On an industrial quote, a plausible number is a wrong number.

Cognitive Learning answers this with hybridisation. The LLM is confined to the interface and to natural-language processing: it captures the request in all its forms, extracts the technical parameters as Smart Tags, and puts the final result back into words. The decision is not its to make. That belongs to a graph of domain and industrial rules encoding your calculation methods, your routings, your compatibilities and your standards. Every value it produces is the result of an explicit path between nodes, and is therefore reproducible and traceable.

The third term matters most over time: this graph updates continuously. When an expert corrects a configuration, adjusts a coefficient or adds a compatibility constraint, they edit the graph itself and the change takes effect immediately. There is no model to retrain and no annotation campaign to run, and our teams need not step in. Your know-how keeps being written after go-live.

COUCHE LANGAGE — LLMComprend le brief, la note vocale, le document, la photo.Rédige la réponse en langue naturelle. Ne calcule rien, ne décide rien.Smart Tags extraitsparamètres techniques structurésDécision + justificationchaque valeur tracée jusqu'à sa règleCOUCHE DÉCISION — GRAPHE MÉTIERVos règles de calcul,compatibilités, gammes,normes et abaques.Déterministe : deux foisla même entrée, deux foisle même résultat.VOS EXPERTSValident, corrigent,enrichissent les règles.Sans réentraînementde modèle, sansintervention technique.
Le modèle de langage n'intervient jamais dans le calcul : il traduit le monde réel en paramètres, puis remet en mots ce que le graphe a décidé.

  • The LLM speaks

    Multimodal understanding of the request and drafting of the answer. No numeric value ever comes from it.

  • The graph decides

    Domain and industrial rules encoded as nodes and relations. Same input, same output, with the decision path attached.

  • Your experts write it

    Every field correction enriches the graph live, with no retraining and no development cycle.

The agentic architecture

Around this foundation, DARYL orchestrates specialised agents: a qualification agent that builds the Smart Tags from the multimodal request, an expert agent that queries the product base and the domain graph, and a learning loop that capitalises on field feedback. All of it on top of a proprietary knowledge graph that structures your company's expertise.

DARYL architecture: multimodal customer request → Qualify AI agent (Smart Tags) → Expert AI agent (product + domain base) → results, under an Intelligent Learning loop

  • Smart Discovery & Selection

    The Smart Tags innovation enables intelligent need discovery and reliable product selection, with continuous learning fed by field feedback.

  • Reliable & Explainable

    Unlike classic generative AI, our proprietary Knowledge Graph guarantees justified, traceable results with no hallucinations.

  • Data Security

    Your data is secured and hosted in France (Microsoft Azure France servers), guaranteeing full confidentiality.

  • Reduced Carbon Footprint

    Our optimised architecture consumes up to 10× fewer tokens, for a more responsible and sustainable AI.

How this technology reaches you

The domain graph isn't built from nothing: it is built from the documents your company already has.

See the deployment