Silo
The problem
Frontier models hallucinate citations, miss relationships between decisions, and offer no audit trail. Lawyers cannot trust answers without evidence.
What is Silo
A 360-degree legal knowledge graph that materializes the legal syllogism (major premise + minor premise = conclusion) into a navigable data structure. It empowers LLMs with structured, relational, and verifiable data about how courts decide.
Numbers
Neo4j Aura, schema v2.1, 9 labels, 15 edge types. 6K+ SAME_AS edges (cross-tribunal entity resolution). Corpus: STJ, TJPR, TJSP, TRF4.
Capabilities
Knowledge Graph
OperationalNeo4j Aura. 9 labels, 15 relationship types. The Subsumption node materializes the legal syllogism. Cross-tribunal entity resolution (SAME_AS). Schema v2.1.
Judge Intelligence
OperationalHow each minister decides by topic. Catalog of decision profiles, cross-judge comparison, profile-to-argumentation conversion. First product with real usage.
Graph-Led Search
FunctionalNeo4j as the primary source, semantic search as fallback. Cross-encoder reranking. Filters by minister, tribunal, date, outcome.
Ingestion Pipeline
Functionalv2.2, 12 stages: ingestion, segmentation, metadata (LLM), normative analysis, factual analysis, canonicalization (regex + embedding + cross-tribunal), Neo4j load. Cross-stage coherence with 4 checks.
Structured Archive
BuildingAgent-friendly layer over petitions. Citation extraction, pipeline prioritization by real lawyer demand. CNJ normalization operational. LLM layer pending validation (GATE).
Legislative Resolution
BuildingLegal reference to canonical ID. Citation parser, alias registry (19 laws), bulk resolution with temporal override. 322 tests, 0 failures. 3 laws in the database (CF, CC, CPC). Local only.
Doctrine
BuildingLocal pipeline: legal books to chunks to embeddings to CLI search. Focus on contract law and civil procedure. Internal supplier for the engine. Pre-release.