A governed framework combining large language models and human oversight for supplier selection

Supplier selection requires organisations to combine structured procurement records with evolving information on sustainability, compliance, operational risk, and market conditions. The ability to reassess suppliers as this information changes is important for maintaining responsive and resilient supply chains. This paper proposes a governed human-in-the-loop framework that connects procurement intent, supplier data, qualitative evidence, multi-criteria analysis, explanation, and managerial review. Language model components support the interpretation of sourcing requests, the extraction of source-linked evidence, and the formulation of revised decision scenarios. Deterministic services remain responsible for weighting, constraint application, scoring, and ranking, while procurement professionals approve consequential analytical and evidential choices. Partial automation of evidence processing and scenario preparation allows supplier evaluations to be updated in response to changing conditions while procurement professionals retain final decision authority. The framework is examined using public procurement records and controlled benchmarks covering scenario formulation, evidence processing, multi-criteria trade-offs, robustness, and explanation fidelity. Its design incorporates schema validation, source-level provenance, explicit evidence policies, deterministic fallbacks, and four human approval checkpoints. Using the extended procurement-fit representation increases Ref@ 10 from 5.259% to 31.642%; the language model components achieve a valid structured-output rate of 1.000, mean evidence-claim F1 of 0.928, and exact preservation of the fixed deterministic ranking in all explanation cases. The study contributes an auditable approach to combining the interpretive capabilities of language models with established supplier-selection methods, providing a basis for more adaptive supplier evaluation while recognising the need for validation in enterprise procurement settings.