Materials AI / Wiki-RAG Position

1. Research axis definition

This axis positions the Atlas itself as the foundation for a future materials AI / Wiki-RAG system for ECC/EGC mixture design, literature reasoning, and experiment planning.

Status:

• Current status: proposal idea / research infrastructure.
• Source memo: ~/Documents/Research_Ideas/NRF_Physical_AI_WikiRAG_Materials_Proposal.md.
• This is not yet a published research output. It should be treated as a strategic proposal direction.

Core logic:

source-grounded ECC Atlas
  -> claim-evidence matrices
    -> material system / mechanism / performance graph
      -> Wiki-RAG retrieval
        -> generative AI-assisted mixture candidates
          -> experimental validation
            -> long-term Physical AI-ready data foundation

2. Global literature anchor

The global anchor is not one material paper but the structured evidence hierarchy of the Atlas:

• Victor Li book: primary conceptual schema.
• Foundational papers: mechanism equations and experimental roots.
• Extension papers: topic-specific evidence.
• Lee lab publications: local experimental and strategic positioning layer.

Relevant Atlas nodes:

• [[micromechanics_based_design]]
• [[strain_hardening_criteria]]
• [[green_ecc]]
• [[geopolymer_ecc]]
• [[extreme_ductility_ecc]]
• [[flaw_design]]
• [[self_healing_ecc]]

3. Lee lab representative papers

This AI/Wiki-RAG axis draws from the entire lab corpus rather than one paper. Representative source clusters:

4. Key evidence and metrics

Current data assets already created:

• 07_visualization/extreme_ductility_extension_claim_evidence_matrix.csv
• 07_visualization/sustainable_ecc_extension_claim_evidence_matrix.csv
• 07_visualization/by_lee_lab_publications_claim_evidence_matrix.csv
• 06_lab_position/by_lee_publication_mapping.csv
• 07_visualization/lab_to_global_lineage_edges.csv

These provide structured rows with paper ID, title, claim, evidence summary, page/section, status, source path, and Atlas node. They are the immediate basis for RAG chunking and query routing.

5. What is distinctive about Lee lab contribution

The lab's advantage is that it has a dense publication corpus across exactly the material dimensions needed for AI-guided ECC/EGC design: binder chemistry, fiber type, recycled fibers, flaw agents, tensile performance, crack width, self-healing, and durability. The Atlas turns this corpus from a publication list into a reusable knowledge graph.

6. Strategic novelty claims

• The Atlas can serve as a source-grounded knowledge substrate for materials AI rather than a generic literature review.
• Wiki-RAG is timely because Physical AI is premature without reliable experimental and literature data infrastructure.
• Claim-evidence matrices allow generative AI outputs to be checked against source-grounded evidence.
• The lab's EGC/ECC corpus provides a realistic testbed for AI-assisted mixture design because it spans binder, fiber, flaw, and durability variables.

7. Manuscript intro/discussion reusable paragraphs

Intro paragraph draft:

Recent advances in generative AI create opportunities for materials design, but cementitious composite development still lacks curated, source-grounded knowledge infrastructures. For ECC/EGC, design decisions depend on coupled mechanisms such as matrix cracking strength, fiber bridging, flaw distribution, and crack-width-controlled durability. A Wiki-RAG based Atlas can organize these mechanisms into a traceable evidence graph, enabling AI-assisted mixture reasoning without disconnecting from experimental sources.

Discussion paragraph draft:

The present Atlas demonstrates that literature, claim evidence, and lab positioning can be transformed into a structured knowledge system. This provides a foundation for future AI-assisted materials design in which proposed mixtures are not generated from ungrounded patterns, but retrieved from and constrained by verified micromechanical and experimental evidence.

8. Proposal background reusable paragraphs

Physical AI for materials development remains premature without a persistent, validated data foundation. A more realistic near-term strategy is to build a Wiki-RAG based materials knowledge system that links literature evidence, experimental data, mixture variables, and performance metrics. Such a system can support generative AI in proposing candidate ECC/EGC mixtures, while preserving traceability to source evidence and enabling iterative experimental validation.

9. Open research opportunities

1. Build a RAG-ready corpus from source notes, paper cards, claim matrices, and lab positioning cards.
2. Define mixture-design metadata: binder, activator, fiber, flaw agent, curing, density, tensile strength, strain capacity, crack width, healing metrics.
3. Develop query templates for mixture recommendation and mechanism explanation.
4. Link AI-generated mixture candidates to required validation tests.
5. Develop a feedback loop from experiment results back into the Atlas.

10. Linked Atlas nodes and source files

• Proposal idea memo: ~/Documents/Research_Ideas/NRF_Physical_AI_WikiRAG_Materials_Proposal.md
• Atlas workflow guide: 09_ai_workflows/atlas_phase1_completion_instruction_guide_2026-08-18.md
• Lab positioning map: 06_lab_position/our_lab_position_map.md
• Lab claim matrix: 07_visualization/by_lee_lab_publications_claim_evidence_matrix.csv
• Global lineage map: 07_visualization/lab_to_global_lineage_map.md
• Future workflow files to create:
• 09_ai_workflows/wiki_rag_ingestion_plan.md
• 09_ai_workflows/atlas_query_templates.md
• 09_ai_workflows/proposal_writing_workflow.md