Skip to content

AI Engine

Fusing the user model and the content model into explainable, diverse recommendations and narratives.

The engine answers one question: given what we know about a visitor and everything we know about the collection, what should they see next, and why? It reads the visitor's interests, gathers candidates by meaning and by expert tags, scores them on several axes, then deliberately adds variety so the result is not ten versions of the same story. Every recommendation keeps a breakdown of why it was chosen, so the system can explain itself.

What you will find here

  • DAC architecture

    The four-layer design (Somatic, Reactive, Adaptive, Contextual) and the meta-control loop.

  • Hybrid scoring

    Combining semantic similarity, Knowledge Graph evidence, and metadata, with bandit-tuned weights.

  • Diversity

    Why pure relevance is not enough, and how MMR re-ranking fights confirmation bias.

  • Explainability

    Making the reason for each recommendation legible, tied to the WP7 ethics framework.

  • Requirements

    The functional and non-functional requirements, traced to visitor user stories.

  • Evaluation

    How the engine is assessed for adaptation, user experience, and ethical sensitivity.

For the implemented, typed codebase behind these concepts, see Implementation: the architecture, the recsys engine, and the code reference.

API for this area

The engine is served by the AI Engine API: search, geo, recommend, narrative, and the recsys ingest/recommend webhook.

The Distributed Adaptive Control architecture of the engine

The engine is organised as a Distributed Adaptive Control (DAC) system: four layers from raw signals up to long-term memory, with a meta-control layer that orchestrates them per request.