PROTEX

Research, teaching and AI evaluation environment for knowledge-based artificial intelligence systems

PROTEX is an independent research and educational programme exploring how knowledge architecture, structured information, and human reasoning processes can shape the design, behaviour, and evaluation of artificial intelligence systems.

The project has developed from a methodological research architecture for structured behavioural case analysis into a broader environment for studying, demonstrating, and teaching how AI systems operate on complex knowledge.

PROTEX includes a working AI prototype that can be demonstrated through a chat-based interface. The system allows users to explore how structured knowledge, retrieval mechanisms, epistemic boundaries, uncertainty, and interpretation affect AI-generated answers.

The system is not designed to generate predictions or automated profiles. It operates as a research and teaching environment for examining how AI systems can support reasoning, decision-making, and analysis while preserving transparency, auditability, and clear evidentiary boundaries.

Research focus

The central focus of PROTEX is the question of how complex knowledge can be represented, retrieved, interpreted, and evaluated inside AI-supported systems.

Narrative and organisational knowledge often combine factual description, interpretation, uncertainty, contextual information, and source-dependent meaning. When processed through conventional retrieval systems, these layers can become blurred, leading to uncontrolled inference or misleading analytical outputs.

PROTEX approaches this challenge by treating knowledge architecture itself as a methodological design problem. The system separates factual, analytical, and interpretative layers so that AI-generated outputs can be examined in relation to their evidentiary basis and confidence level.

This makes PROTEX useful not only as a research system, but also as a teaching environment for demonstrating why AI evaluation cannot be reduced to checking whether an answer is simply correct or incorrect.

Architecture and methodology

At the architectural level, PROTEX combines structured knowledge representation, semantic retrieval, deterministic metadata filtering, explicit query routing, and controlled response generation.

The current prototype operates through three functional modes: factual, analytical, and interpretative. This allows different types of questions to be handled under different epistemic conditions, depending on whether the system is expected to retrieve facts, support analysis, or engage with interpretation.

Cases are represented as structured analytical objects composed of case profiles, narrative fragments, behavioural descriptions, contextual metadata, and indicators of uncertainty. This design enables controlled comparison of complex information while maintaining traceability between system outputs and the underlying source material.

The objective is not to automate explanation or replace human judgement, but to create an environment in which human users can observe, question, and evaluate how AI systems operate on structured knowledge.

Teaching and demonstration value

PROTEX functions as a practical teaching demonstrator for students, professionals, and organisations interested in understanding how modern AI systems are designed, evaluated, and applied.

Because the system can be shown through a live chat interface, it allows participants to observe how questions are routed, how answers are grounded in source material, how uncertainty appears, and how different knowledge structures affect AI behaviour.

The behavioural case domain was selected not for sensational purposes, but because it provides a demanding test environment for AI systems. The material contains ambiguous evidence, multiple interpretations, fragmented sources, uncertainty, and different levels of confidence. This makes it particularly useful for demonstrating how AI systems should handle complex knowledge responsibly.

In educational settings, PROTEX can support lectures, workshops, seminars, and practical demonstrations on knowledge architecture, Human-AI Collaboration, AI evaluation, RAG systems, AI governance, and decision-support environments.

AI evaluation and benchmark research

PROTEX also provides the foundation for an ongoing AI evaluation programme. The project is used to design and test benchmark methodologies for assessing the quality, reliability, and trustworthiness of AI-generated answers.

This work treats benchmarking as more than a simple comparison between questions and correct answers. It examines factual grounding, source traceability, completeness, reasoning boundaries, uncertainty handling, interpretation control, and the stability of answers across different AI systems.

The PROTEX benchmark research shows how evaluation can be used not only to measure AI performance, but also to teach students and organisations how to think critically about AI outputs, knowledge quality, and the limits of automated reasoning.

This makes PROTEX a bridge between research, education, and applied AI assurance.

Applied AI and decision-support environments

Beyond the behavioural case domain, PROTEX explores broader applications of knowledge architecture and AI evaluation in organisational settings.

The same design principles can be applied to employee onboarding systems, operational knowledge assistants, product knowledge environments, procedural AI assistants, and decision-support tools operating on organisational knowledge.

This broader direction allows PROTEX to connect theoretical research with practical system design, showing how structured knowledge can improve the usefulness, transparency, and reliability of AI systems in real environments.

The long-term objective is to develop a coherent framework for designing, teaching, and evaluating AI systems that support human decision-making without removing human responsibility.

Ethics and design principles

– research and educational use
– historical cases documented in public sources
– no operational investigative use
– no automated profiling or diagnosis
– no replacement of human judgement
– explicit modelling of uncertainty and missing data
– separation of factual, analytical, and interpretative layers
– transparent and auditable system architecture

Contact

Email
karol@protex-profiler.ai