In industry, mastering the expert language is a prerequisite for automating high-value tasks: drafting technical quotes, assisting diagnostics, or selecting complex solutions. To deliver on that, DARYL embeds a new technology: Advanced KAG, blending RAG with a semantic knowledge graph.
The technical concept: RAG + Knowledge Graph
RAG (Retrieval-Augmented Generation) combines a retrieval module that queries a document base with a generative module that produces an answer based on those documents. But in an industrial context, this approach quickly hits its limits: ambiguous domain terminology, sometimes implicit technical documents, an inability to identify what matters in a project.
This is where Advanced KAG comes in. It adds a domain knowledge graph between the retrieval engine and the generative model. The graph contains:
- •A domain ontology (relationships between products, uses, standards, constraints)
- •Validated technical entities (normalised terms, units, categories)
- •Reasoning rules (conditions of application, dependencies, compatibilities)
The generative model is therefore guided not only by texts but also by an explicit knowledge structure. It can “reason” in the context of the domain.
Classic RAG vs Advanced KAG: two levels of expertise
Example 1 — Choosing a pressure regulator
« The RX400 is a regulator suited to high-pressure hydraulic circuits. »
« The RX400 is recommended because it is compatible with flow rates above 60 l/min in circuits using HLP46 fluid, with operating temperatures up to 80 °C. It complies with ISO 1219-1, required in your sector. »
Example 2 — Energy-performance gap in a building
« Gaps can be linked to insulation, ventilation or user behaviour. »
« The 32% gap observed mainly comes from replacing the single-flow MVHR with a double-flow system without recalibrating the minimum air flow required for that dwelling type (RT2012, zone H1a), identified via the knowledge graph cross-referencing thermal and regulatory data. »
AI that learns from field experts
What makes Advanced KAG alive is the collaborative field-validation loop built into DARYL. Experts annotate, correct or complete generated answers in real situations. The knowledge graph enriches itself automatically: new synonyms, emerging domain constraints, alternative logics.
For example, a technician may report that, in the field, a particular sensor model fails in high-humidity zones — even if it isn't documented. That information becomes a domain rule in the graph. The result: on the next quote in a similar environment, DARYL will avoid that model.
Truly specialised AI
Advanced KAG technology blends the power of generative language with the rigour of structured knowledge. Thanks to it, DARYL is no longer a generic assistant: it becomes an embedded expert, understanding the domain language, the project constraints, and continuously learning alongside field teams.
