Victor P Unda

Designed an evidence-first validation framework for question answering systems, introducing Meaning–Utility Estimation (MUE) and Diversity–Utility Estimation (DUE) for deterministic evaluation and selection of evidence units. Built and deployed the Victor Evidence API prototype, an evidence-gated system that returns an answer only when explicit textual support is present under fixed admissibility conditions (otherwise returns a controlled “no answer” result). Developed a deployment-focused follow-up study comparing observed behavior across two pipelines: corpus-based candidate retrieval and externally supplied / generated candidate-unit evaluation. Research areas Evidence-first validation for QA systems Deterministic evidence gating and traceability Candidate-unit evaluation (corpus retrieval and generated/supplied units) Meaning–Utility Estimation (MUE) Diversity–Utility Estimation (DUE)