Dev.to WebDev πŸ›  Dev πŸ‘ 0 πŸ“– 1 min read

Building a 22-Market Cosmetic INCI & Regulatory Knowledge Graph with LLM Deep Research Pipelines

When analyzing cross-border skincare formulations and regional regulatory compliance, raw ingredient lists (INCI) and fragmented market entry rules are notoriously difficult to standardize across different jurisdictions.

When analyzing cross-border skincare formulations and regional regulatory compliance, raw ingredient lists (INCI) and fragmented market entry rules are notoriously difficult to standardize across different jurisdictions.

Recently, I built an automated research and data-indexing pipelineβ€”leveraging Claude & Gemini Deep Research architecturesβ€”to structure formulation ontologies, clinical benchmark metrics, and regulatory readiness across 22 global markets (spanning North America, APAC, 7 newly activated European countries, and 6 Latin American markets).

Full interactive atlas and structured datasets are live at RIMAN.WIN β€” Clinical Science & 22-Market Growth Atlas.

1. Data Architecture: From Raw INCI to Multi-Market Ontology

Instead of treating product pages as unstructured text, the pipeline normalizes every formulation into three relational JSON layers:

  • Descending INCI Ingredient Ontology: Mapping exact botanical extracts (such as Giant BYoungPoolβ„’ and Araliadiol), Jeju lava seawater minerals, and microfluidizer delivery systems against CosIng/PubChem standards.
  • Climate Γ— Skin-Type Stress Matrix: Evaluating barrier recovery and trans-epidermal water loss (TEWL) behavior across contrasting microclimatesβ€”from cold-dry European winters (EU CPNP compliance) to high-humidity APAC monsoons.
  • 22-Market Regulatory & Entry Index: Structuring local currency frameworks, registration pathways, and cross-border distribution rules into machine-readable schemas.

2. Programmatic Multilingual Static Rendering

To ensure zero-latency global access and clean semantic indexing across 11 languages (en, zh-TW, zh-CN, ja, ko, es, fr, de, etc.), the frontend compiles these JSON ontologies into static edge-cached routes with bidirectional hreflang and JSON-LD structured metadata.

3. Open-Science Publishing & Reproducibility

To keep the underlying formulation and market readiness benchmarks verifiable for fellow developers and data researchers, subsets of the structured schemas and multi-scenario research reports are continuously archived with persistent DOIs across open repositories (GitHub, Zenodo, Harvard Dataverse, and Hugging Face).

  • Interactive 22-Market Biome & Formulation Atlas: https://riman.win
  • Topics Covered: Descending INCI matrices, P&K clinical benchmarks, and cross-border regulatory mapping.

Feedback on multi-agent data extraction pipelines or programmatic multilingual SSG architectures is always welcome in the comments!

πŸ“° Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.