From Knowledge to Inference: Formalizing Specialized Public Health Reasoning on GlobalHealthAtlas

Public health reasoning requires population level inference grounded in scientific evidence, expert consensus, and safety constraints. However, it remains underexplored as a structured machine learning problem with limited supervised signals and benchmarks. We introduce GlobalHealthAtlas, a large scale multilingual dataset of 280,210 instances spanning 15 public health domains and 17 languages. We further propose a large language model (LLM) assisted construction and quality control pipeline with retrieval, deduplication, evidence grounding checks, and label validation to improve consistency at scale. Finally, we present a domain aligned evaluator distilled from high confidence judgments of diverse LLMs to assess outputs along six dimensions: Accuracy, Reasoning, Completeness, Consensus Alignment, Terminology Norms, and Insightfulness. Together, these contributions enable reproducible training and evaluation of LLMs for safety critical public health reasoning beyond conventional QA benchmarks. We publicly release project codebase, evaluator, and model at:: https://github.com/Jan8217/GlobalHealthAtlas, https://huggingface.co/aerovane0/GlobalHealthAtlasPublicEvaluator and https://huggingface.co/aerovane0/GlobalHealthAtlasPublicModel

As AI systems are increasingly used in healthcare, it is crucial that their advice is safe, scientifically accurate, and reliable. However, we currently lack the proper tools and data to teach and test AI on complex public health issues. To solve this, we created GlobalHealthAtlas, a massive, multi-language database containing hundreds of thousands of verified public health examples. We also built an automated grading system that acts like an examiner, scoring AI responses on things like accuracy, logic, and safety. Ultimately, our work provides the essential training materials and testing standards needed to build safer and more trustworthy AI assistants for public health.