The EPS Research Astro-RAG Platform preprint (arXiv:2605.30384, May 2026) assembles four machine-readable corpora totaling 772 objects spanning z=0 to z~5.68, linked by the omega kinematic correction from Flynn & Cannaliato (2025). This yields a sign reversal in boundary-point angular velocity—from SPARC mean +7.06 rad/Gyr locally to Z1 median -13.05 rad/Gyr at high redshift—while bundling 120 Jupyter notebooks and a high-school track under one CC-BY-4.0 release. Unlike piecemeal tool announcements, the platform treats LLMs, archives, and pedagogy as interdependent layers, directly addressing the fragmentation that still dominates coverage of AI in astronomy. It preserves full SPARC fidelity (Lelli et al. 2016, ApJS, 175 galaxies with photometric decompositions) and augments it with THINGS, LITTLE THINGS, WALLABY DR2, and ALMA [CII] data, yet the work remains a preprint without independent peer review. Limitations include reliance on existing heterogeneous surveys, absence of new observations, and dependence on future RAMSES simulations to test the omega reversal. Synthesizing this with Lewis et al. (2020) on retrieval-augmented generation shows how Astro-RAG operationalizes RAG at dataset scale rather than query scale, creating reproducible workflows that isolated LLM demos rarely achieve. The result is not incremental tooling but an early template for research ecosystems where data, models, and training co-evolve.