Towards Precision Environmental Health: AERIS, a Research Agent for Air-Pollution Exposure Assessment

Air pollution harms human health and contributes to millions of deaths each year. Yet many public-health studies fail to account for it, despite recording locations and times that enable exposure reconstruction. Producing an analysis-ready variable requires suitable air-quality data, spatiotemporal alignment, coverage checks, and a record of each decision. We present AERIS (Airborne Exposure Research & Investigation System), a research system that lets Codex, Claude Code, and other command-line agents estimate personal and population-level air-pollution exposure. AERIS accepts inputs at different spatial and temporal resolutions, from ZIP codes to mobility trajectories and administrative areas. It queries public air-quality repositories including OpenAQ, computes exposure, and preserves the evidence needed to review each result. To evaluate its performance and practical value in a real-world setting, we conducted a large-scale replication study using source-substitution reanalysis: AERIS reconstructed exposure from independent public air-quality data and reran published analyses with available public health-outcome data. We screened 18,919 papers, identified 14 eligible for attempted reanalysis, and produced fitted estimates for nine. AERIS and published effect ratios differed by a median factor of 1.04 (median absolute log-ratio difference, 0.040). The estimates were strongly correlated (Pearson (r=0.834), 95% CI 0.314–0.969). Together, these results show that AERIS can construct useful exposure records across varied settings and preserve a reviewable path from public air-quality data to health-effect estimates, enabling research agents to autonomously incorporate air-pollution exposure into public-health studies.