AI-Powered Knowledge Engines as Research Infrastructure for Systematic Knowledge Discovery

This vision paper proposes knowledge engines as a framework for how intelligent systems systematically discover, integrate, and generate knowledge. We define nine integrated capabilities and present CopernicusAI as a working implementation: 59,499 indexed research papers (100% dense-embedded), 594 JSON-canonical process diagrams across six discipline families, vector search, and retrieval-augmented generation. Empirical content (June 2026 update): A preliminary retrieval pilot over a frozen corpus of 59,499 documents (30 queries) compares lexical TF-IDF (mean nDCG@10 = 0.545) with OpenAI dense retrieval using text-embedding-3-small (mean nDCG@10 = 0.828) under a single-assessor protocol with an expanded 558-pair judgment pool. This record contains: the manuscript PDF; the May 2026 evaluation freeze (frozen corpus research_papers_20260526.jsonl.gz, Colab notebook, lexical rankings, query list); and the June 2026 dense track A bundle (dense_track_a_bundle_20260608.zip: cached corpus embeddings, dense rankings, expanded judgments, metrics manifest, reproduction script). SHA-256 checksums are in manifest.json.