Research grounded
in real operations.
My research covers enterprise AI, industrial knowledge systems and operations management. It starts from what real deployments deal with: scattered data, old infrastructure, governance rules and decisions with physical consequences.
Research areas
Papers
Papers published elsewhere link to the publisher. For work I can share myself, leave your email and the PDF is sent to you.
Robotic Process Automation: A Qualitative Journey Through RPA's Impacts on Company Employees
abstract
Robotic process automation is usually justified in headcount terms, which is the least interesting thing about it. This study takes a qualitative route instead: thirteen semi-structured interviews with RPA practitioners across four multinational companies, asking what automation actually did to the people around it. The finding is that RPA does not straightforwardly displace employees, but it does reshape their work and the governance around it — and how much it reshapes depends on which automation strategy the organisation chose. Task-oriented programmes lean on citizen developers and move quickly at the cost of a coherent process view; process-oriented programmes rely on professional developers and demand far more structured governance. The chapter draws these threads into an integrated framework linking automation strategy, governance model, upskilling, and employee adaptation.
authors
Edgar Simões, Ana Correia Simões, José Coelho Rodrigues, Pedro Miguel Lourenço
key contributions
Reconfigurable Smart Production System for Prefabricated Panelised Construction
abstract
Prefabricated panelised construction offers significant efficiency gains but requires production systems that can adapt to multiple product variants without costly process redesign. Traditional fixed-sequence production lines struggle with design variability, leading to bottlenecks and rework. This paper presents a reconfigurable production architecture that maintains production coherence across variant design spaces while coordinating fabrication sequences, logistics constraints, and multi-site assembly workflows. The approach is grounded in a real industrial deployment and demonstrates how structured ontology-driven production planning can enable factory-level flexibility without sacrificing traceability or quality control.
authors
Pedro Miguel Lourenço
key contributions
Essays
Essays work through a single problem in the space between a research note and a full paper.
File-Based Knowledge Graphs and Retrieval-Augmented AI for Complex Project Delivery
abstract
Complex engineering and transformation projects generate heterogeneous knowledge that is difficult to integrate, trace, and reuse across multi-year lifecycles. Although knowledge graphs (KGs) and retrieval-augmented generation (RAG) have independently matured, many teams still lack a practical path from fragmented documents to explainable AI-assisted decision support. This paper develops an end-to-end design pattern for file-based project knowledge graphs: KGs whose canonical representation resides in structured files (Markdown + YAML + explicit links), rather than in dedicated graph databases. It is based on a prototype built for a real-case, multi-year, multi-site industrial MES implementation project, and presents a detailed design that covers ontology governance, graph encoding patterns, agentic retrieval loops, provenance rules, human-in-the-loop write controls, and production-oriented cost optimisation.
authors
Pedro Miguel Lourenço
key contributions
Beyond Speech-to-Text: Preserving Meaning in Clinical Transcription
abstract
This research essay examines AI-assisted clinical transcription as a problem of meaning preservation rather than speech recognition alone. A transcript can read perfectly and still be clinically wrong: "no fracture" becomes "fracture", a decimal shifts, left becomes right, a small finding disappears. The essay sets out what has to survive the journey from dictation to report — findings and negation, anatomy and laterality, numbers and units, uncertainty, and report structure — and proposes a human-in-the-loop workflow in which AI prepares and checks a draft while the physician keeps final validation and release authority. It develops a controlled semantic post-processing layer that corrects recurring ASR errors where the source supports the correction and flags them where it does not, a risk-based review taxonomy that directs attention without letting the system decide clinical truth, and the surrounding platform — workflow states, versioned report models, audit trail — that makes any of it accountable. The work combines solution design with early qualitative exploration; it is explicitly not a completed clinical-validation study, and it states where that boundary falls.
authors
Pedro Miguel Lourenço
key contributions
Ontology Alignment Patterns for Manufacturing ERP Integration
abstract
Every enterprise ontology project eventually collides with the ERP. The data is there, the semantics are not, and a full schema migration is never on the table. This essay works through the practical design patterns for mapping ERP data structures onto a domain ontology without rewriting either — where to put the mapping layer, how to version it as the ERP configuration drifts, and which classes of semantic mismatch are worth modelling versus worth ignoring.
authors
Pedro Miguel Lourenço
key contributions
Human-AI Decision Handoff Models in Industrial Operations
abstract
When should an agent escalate to a person? Most deployments answer this with a confidence threshold, which is the wrong instrument — confidence measures the model's certainty, not the cost of being wrong. This essay formalises the conditions, triggers, and interface patterns for reliable human-in-the-loop operation in settings where a bad automated decision has physical consequences, and argues for handoff rules grounded in consequence and reversibility rather than model self-report.
authors
Pedro Miguel Lourenço
key contributions
Cost Optimisation Strategies for Enterprise RAG Pipelines
abstract
A retrieval pipeline that is correct but uneconomical does not reach production. This essay collects the techniques that reduce inference cost in production RAG systems without giving up retrieval quality or explainability: routing by task rather than by default, caching at the semantic layer instead of the prompt layer, structuring retrieval so that the expensive model sees less but better context, and measuring the whole thing in cost-per-answered-question rather than cost-per-token.
authors
Pedro Miguel Lourenço
key contributions
How the research is done
Primary research
Original design patterns, prototypes, and frameworks developed from real project contexts — not retrospective literature reviews.
Applied prototyping
Ideas are validated through working prototypes built on real operational environments, not toy examples.
Industry validation
Findings are tested where they are hardest to defend: inside live programmes, under audit requirements, against data nobody cleaned first.
Open publication
I publish what I find and make it available to the wider community of practitioners and researchers.
Want to work on this together?
I'm open to research partnerships with companies and universities working on the same problems.