This page brings together selected system directions developed within the PROTEX programme.

The work focuses on how AI systems can operate on structured knowledge, under explicit constraints, within research, educational, and organisational environments.

Rather than treating AI as a general conversational layer, these systems are designed around bounded roles, controlled access to knowledge, clear responsibility structures, and transparent evaluation of system behaviour.

Research systems. Teaching demonstrators. Applied AI environments.

Across all system directions, the main concern is how knowledge is structured for AI systems, how access to that knowledge is controlled, and how system behaviour changes when AI is embedded into real workflows, procedures, learning environments, and decision contexts.

The systems developed within PROTEX are not only technical prototypes. They also function as demonstrators for explaining how AI systems retrieve, interpret, organise, and evaluate knowledge.

This makes them useful in research, workshops, lectures, and organisational learning, where participants can observe how AI behaviour depends on knowledge architecture, retrieval design, uncertainty handling, and human oversight.

This includes role-specific systems, procedural guidance layers, expert knowledge interfaces, structured retrieval architectures, and live demonstrators for AI evaluation.

The PROTEX research prototype is a working AI environment available through a chat-based interface.

It demonstrates how structured behavioural and narrative knowledge can be queried, analysed, and interpreted under controlled epistemic conditions.

The prototype currently operates through three functional modes: factual, analytical, and interpretative. This allows users to observe how different types of questions require different response boundaries and different levels of confidence.

As a teaching demonstrator, the prototype can be used to show how AI systems handle ambiguity, source hierarchy, uncertainty, factual grounding, interpretation, and evidence-based response generation.

PROTEX also serves as the foundation for AI evaluation and benchmark research.

The benchmark work developed within the project examines not only whether an AI answer is correct, but also whether it is grounded in evidence, complete, stable, traceable, appropriately cautious, and resistant to uncontrolled inference.

This direction is used to study how modern AI systems behave when operating on structured knowledge repositories and how evaluation methods can be designed for real organisational knowledge environments.

This makes AI evaluation a practical research method, an applied assurance tool, and a teaching framework for developing critical understanding of AI-generated outputs.

Procedural AI systems are explored as bounded support systems for environments in which people rely on documented procedures, repeatable operational logic, and organisational knowledge.

These systems are not designed to decide on behalf of users. They retrieve, organise, and explain approved procedural knowledge, supporting action within defined organisational boundaries.

They are particularly relevant for onboarding, operational decision support, internal knowledge assistants, and procedural learning environments.

Expert AI systems are studied as role-specific knowledge interfaces built around verified products, services, professional knowledge, and domain rules.

The purpose of these systems is structured explanation of complex knowledge within explicit response boundaries.

They are designed to support users by making expert knowledge easier to access, without removing the need for human judgement, contextual awareness, and responsibility.

A central part of the programme focuses on knowledge backends, retrieval layers, metadata design, and ingestion pipelines for AI systems.

Many AI system failures originate not from the model itself, but from poorly structured, incomplete, ambiguous, or uncontrolled knowledge.

PROTEX examines how knowledge can be modelled so that AI systems are able to retrieve relevant information, preserve context, distinguish between factual and interpretative layers, and avoid unsupported inference.

This work focuses on structured modelling, metadata preservation, controlled retrieval, deterministic knowledge access layers, and epistemic constraints.

The system design blueprint is treated as a framework for thinking about AI systems before they influence real decisions, learning processes, or organisational workflows.

It examines how systems interact with knowledge, procedures, uncertainty, responsibility structures, and human decision-making.

The blueprint is used to ask where an AI system's role should begin, where it should end, and how its behaviour can be evaluated before deployment.

Governance is approached as a question of responsibility, decision authority, knowledge ownership, and risk boundaries.

Within PROTEX, governance is not treated only as a policy layer. It is treated as something that should be reflected in system architecture, retrieval control, response boundaries, and evaluation methodology.

This includes responsibility mapping, knowledge ownership, human oversight, uncertainty handling, and system-level enforcement.

• behavioural case analysis and criminological knowledge environments
• hospitality and operational decision-support systems
• onboarding and organisational knowledge assistants
• expert advisory and product knowledge systems
• educational demonstrations of AI behaviour and evaluation

These domains are used to observe how system behaviour changes under different knowledge structures, decision constraints, levels of uncertainty, and user responsibilities.

Email
karol@protex-profiler.ai