Gulger Mallik

Integrating Machine Learning and Knowledge Graphs for Explainable Soil Organic Carbon Assessment in Agroforestry Systems

Reliable assessment of soil organic carbon is essential for carbon removal, agroforestry planning, and credible Measurement, Reporting and Verification (MRV) systems. However, existing approaches face persistent challenges related to heterogeneous soil data, inconsistent sampling depths, limited spatial coverage, and the low interpretability of data-driven prediction models. This paper proposes a hybrid artificial intelligence architecture that integrates supervised machine learning with graph-based knowledge models to support explainable soil organic carbon assessment in agroforestry systems. The approach combines a data harmonisation pipeline, Random Forest models for estimating organic carbon concentration and carbon stocks, a graph neural network embedding model for identifying spatially coherent soil archetypes, and a Bayesian-network-based explanatory graph for interpreting environmental profiles associated with high and low organic carbon conditions. The framework is evaluated through a Mexico-focused case study using the WoSIS soil profile database enriched with climate, land-cover, ecoregion, and expert-recommended soil covariates. Results show that Random Forest models provide moderate predictive performance, with R2 values ranging from 0.38 to 0.49, while the graph-based component reveals geographically meaningful soil clusters and supports interpretable profiling of organic carbon conditions. The findings indicate that hybrid AI architectures can support soil carbon assessment by combining quantitative prediction with spatial contextualisation and explainability. The paper also discusses limitations related to data availability, geographic generalisability, and the need for further validation before operational MRV deployment.

Status: Under Review

Bridging Organisational Data Silos through Multi-Agent Generative Systems and Integrated Relational Governance

Modern organisational leadership is currently navigating a systemic shift from intuitive, gut-based pro cesses toward a more rigorous, evidence-based decision-making paradigm. This transition is often defined by Dual Process Theory as a move from 'Type 1' thinking, which is fast and automatic to 'Type 2' thinking, a slower and more analytical approach that relies on empirical data. Despite the professionalisation of management, many organisations remain hindered by data silos, isolated repositories of information that fragment knowledge and prevent a 'single source of truth'. This paper presents a comprehensive review of three primary archetypes designed to overcome these barriers: Multi-Criteria Decision Analysis (MCDA) for discrete, high-stakes choices; Decision Intelligence and Business Intelligence (DI/BI) for continuous operational optimisation; and Stakeholder Relationship Management (SRM) for maintaining the 'Social Licence to Operate'.The review argues that these archetypes are increasingly interdependent, leading to an emergent paradigm termed Hybrid Decision Intelligence (HDI). In this integrated model, qualitative human insights from SRM flow into the mathematical frameworks of MCDA, while real-time evidence from DI/BI platforms continuously updates the decision-making base. To manage this complex convergence, the paper explores the use of Multi-Agent Generative Systems (MAGS), where specialised AI agents collaborate to monitor sentiment, verify compliance, and suggest strategic actions. By implementing a tiered autonomy framework, organisations can delegate routine tasks to autonomous agents while ensuring that high-stakes, value-laden decisions remain under human oversight. Ultimately, this synthesis offers a roadmap for build ing resilient, data-driven organisations capable of navigating the deep uncertainties of the modern global landscape

Status: Under Review

ATMAS: Design, Implementation, and Evaluation of an Automated Test Management and Asset Scheduling System for a University Research Engineering Laboratory

ATMAS presents an automated platform for test management and asset scheduling, replacing spreadsheet-driven workflows with a traceable and operationally efficient system.

Status: Manuscript in preparation