Research

How AI should reason, ask first and remember.

Open technical reports by Argjend Bajraktari on context-aware AI in enterprise systems and on professional memory. Each proposes an architecture and a way to evaluate it.

Enterprise AI

RATIO

Version 1.0 · 3 November 2026

An Architectural Study of Context-Aware Artificial Intelligence in Enterprise Systems

How can an AI system use operational context while respecting permissions and keeping its decisions open to review? RATIO proposes a reference architecture that separates model-assisted interpretation from identity, permissions, evidence, approval and controlled execution, with a synthetic worked example and an evaluation protocol.

Published
Zenodo, 3 November 2026
DOI
10.5281/zenodo.23911976
Licence
CC BY 4.0
Peer review
Not peer reviewed

Professional memory

LIBRA

Version 1.0 · 5 November 2026

An Evidence-Centred Architecture for Professional Memory and Decision Continuity

How can professional knowledge keep its sources and reasoning as documents, teams and decisions change? LIBRA proposes an evidence-centred architecture that connects source findings, evidence dates and human review, and a way to evaluate it.

Published
Zenodo, 5 November 2026
DOI
10.5281/zenodo.23986865
Licence
CC BY 4.0
Peer review
Not peer reviewed

How to read them

Read them as designs, not results.

The reports describe how such systems could be built and tested.

What they are

Technical reports that propose architectures, worked examples and evaluation methods. Both are version 1.0, open access and published on Zenodo under CC BY 4.0.

What they are not

They are not peer reviewed, and they report no completed comparative experiments or measured performance results. Neither describes the performance of Ancilor or of any installation.

Who wrote them

Argjend Bajraktari, ORCID 0009-0009-6135-7509. The author has a commercial interest through Harrimont; the reports present a practitioner perspective, not an independent evaluation.

In practice

Where the research meets the work.

In Ancilor, AI helps people review and summarise. It does not change records, and a person makes the decision. The reports examine how far that separation can be designed, tested and kept.