RoleProject Lead (oversight)
TagsGenAI · LLMs · Content Automation · RAG
TimelineNov 2023 — Feb 2025

Blog generation, grounded in research

The core of the system is blog generation. The pipeline takes a human-provided list of microhabits and uses an LLM to expand each one into a set of candidate blog topics. From there it branches by content type: for research-backed blogs, the pipeline searches out the top research papers for a given topic and feeds excerpts from those papers in as context, so the writing is grounded in real science rather than model recall; for lighter pieces, the LLM draws on its own knowledge base. Grounding the science-based blogs in sourced papers is what keeps the content credible for a health-and-wellness audience, where accuracy matters.

Full-package blogs & multi-channel fan-out

Each blog ships as a complete package rather than just text. Every piece comes with a Midjourney-generated image and either LLM- or programmatically-generated SEO parameters; the research-based blogs additionally carry a glossary — generated by the LLM and cross-referenced against an online neuroscience dictionary wherever possible — plus a citation for the source paper.

To keep a human in the loop, we built a minimal Google Workspace-based approval workflow where a reviewer can edit, comment, and approve or reject each blog and its artifacts before anything goes out. Finally, approved blogs are repurposed into other editorial formats — email newsletters, podcasts, and social media posts — so a single source blog fans out across the client's marketing channels.