Trust at the Seams: Open-Source, AI & Ontology-Driven Polyglot Persistence Layer for Industrial

Industrial AI must reason across incompatible data shapes — relational, graph, vector, time-series, search, object storage, and blockchain provenance — yet most RAG/agentic stacks still bolt one vector store onto one source of truth. This talk introduces Veritas, an open-source, ontology-driven polyglot-persistence platform built for real autonomous-wind-farm structural-health-monitoring pipelines. It will cover: (1) a trust-scored routing layer attaching uncertainty-aware confidence bounds to every result, so engineers know HOW MUCH to trust an answer, not just what it says; (2) a GraphRAG architecture for noisy industrial SCADA/document corpora, with a multistep layout-analysis and semantic-chunking pipeline; (3) an agentic data-quality framework instrumented with OpenTelemetry for non-deterministic LLM workflows; and (4) a UUID provenance system anchored across stores for tamper-evident audits. Attendees leave with open-source patterns for combining knowledge graphs, vector databases, and blockchain provenance in production, plus a reference architecture adaptable to their own data estates.