An alternative to knowledge graphs for storing loosely structured content (fleetingswallow.com)

AILang announced the Knowledge Amalgamator, an alternative to one-size-fits-all knowledge graphs that makes heterogeneous documentation AI-accessible by processing sources according to their inherent structure and organizing extracted content around a lightweight Person-style memory entity (“the Archivist”). Rather than forcing every document into entity–relationship triples, the system classifies sources as well‑structured, semi‑structured, loosely‑structured, or ambiguous, skims and indexes structured docs (outlines/anchors), and heavily internalizes conversational or fragmented sources to extract decisions, risks, procedures and cross-references. This mirrors human information management and solves common problems—ambiguity, missing relationships, context sprawl—without duplicating or losing the original document’s organizational clarity. Technically, ingestion and access are split into two deterministic phases: knowledge_amalgamator builds exactly six provenance-rich memory files (episodic_memory.jsonl; semantic_memory.graph.json; procedural_memory.json; outline_index.json; citation_index.csv; manifest.json), and knowledge_responder consults those files plus original sources to answer conversational queries with precise citations and no hallucinations. The Person entity unifies episodic, semantic, procedural and outline memories for natural query patterns, while structure-aware rules (not large-scale model training) keep costs and complexity low. For AI/ML teams, this offers a pragmatic, auditable, and maintainable path to grounding LLMs in messy enterprise data without heavy graph engineering or expensive domain model training.