Insights on AI transparency, knowledge management, and building trustworthy enterprise systems.
GlassBox measures contextual precision, semantic faithfulness, and structural fidelity for any RAG system — then commits a tamper-proof trust scorecard to a Hyperledger Fabric ledger.
A point-and-click UI replacing the curl-only workflow for editing per-space normalization rules — aliases, label maps, predicate maps, and CSV field maps — with dry-run validation and a full round-trip contract.
Every answer Anatypical generates is anchored to specific document passages and entities via persistent Neo4j graph edges — surviving re-ingestion, entity merges, and session restarts.
The perplexity gate decides whether a query needs document retrieval — using either a structured LLM classifier or token log-probabilities — before routing to the retrieval pipeline.
Barnyard's Memify phase groups related TextNodes into LLM-generated TopicClusters — semantic themes that improve retrieval recall and power the knowledge graph canvas view.
Barnyard stores conversation turns as a persistent graph in Neo4j, enabling durable context, branching threads, and provenance tracking that survives session restarts.
A deep dive into Barnyard's two-phase pipeline: GLiNER for zero-shot NER and a single-pass LLM for relation triplets, followed by cross-document deduplication and pre-materialized RelationStar summaries.
How Barnyard normalizes inconsistent LLM-extracted predicates into a 30+ canonical predicate ontology across 13 semantic groups, preventing knowledge graph fragmentation.
Barnyard combines entity search (BM25 + vector), topic cluster retrieval, and knowledge graph expansion into a single ranked passage pool using Maximum Marginal Relevance fusion.
Traditional RAG forces a choice between semantic search and keyword matching. Hybrid RAG eliminates that tradeoff.
