NeuroLab AI

NeuroLab AI is a scientific research prototype that helps researchers investigate drought-related patterns in published Ghana maize trial data. It connects a research objective and reviewable analysis plan to bounded Meta-guided tool execution. Researchers can inspect genotype rankings, water-regime comparisons, trait associations and environment sensitivity, then choose a follow-up analysis or stop when the data cannot support a stronger conclusion. The project addresses a practical problem: a convincing AI narrative can be difficult to distinguish from an analysis grounded in measurements. NeuroLab AI retains source lineage, model decisions, computed artifacts and scientific limitations in saved runs and exportable reports. Charts and cinematic replay reflect actual recorded results and events; deterministic mode is explicitly labelled and makes no model calls. On October 29, a judge-account investigation completed using the required Baseten-hosted Meta model, meta-models/muse-spark-open-1026, with six real model selections, six scientific tools and six result artifacts, without recovery. An earlier exact-model attempt rejected one invalid model response, used a labelled recovery and remained PARTIAL; both attempts are retained. The current Ghana dataset contains 864 plot records covering 72 genotypes. Findings are exploratory, water regimes are confounded with environments, and the application does not provide validated field recommendations. Judges can run the application locally without hosted credentials or paid API access by following the “Judges can run locally without hosted access” section of the README. This path uses the published Ghana dataset and real scientific tools in explicitly labelled deterministic mode. The hosted Meta-assisted workflow has separate recorded execution evidence. A tested hosted judge account is also available, with credentials to be supplied through an organizer-approved private channel.