EDI modernization
Map partner-specific transaction sets and implementation guides to canonical models while preserving qualifier, loop, and code-set context.
Products / ARI
Ari helps organizations infer, explain, and govern mappings between complex data shapes.
Automated Relation Inference
Complex mappings are usually built by hand. Experts compare two schemas, reconcile naming and structural differences, encode special cases, and repeat the work as systems and partner formats evolve.
Ari treats mapping as a semantic inference problem. It combines structure, names, types, qualifiers, code sets, ontology bindings, constraints, and prior mappings to propose relationships between two data shapes. Every proposal can be scored, explained, reviewed, and reused.
The result is faster onboarding, more consistent mapping decisions, and a governed body of integration knowledge that grows with each project.
Schema Matching Studio
Ari's studio brings shape compilation, relation inference, and human review into one workflow. Inspect compiled structures in the Shape Graph Editor, then review inferred mappings, evidence, and coverage in the Relation Editor.
Shape Graph Matching
Ari represents each side of a mapping problem as a semantic shape graph. A shape can describe an EDI transaction, an API or event contract, a canonical domain model, a file, a document-derived record, or an internal schema.
Rather than flattening those formats into lists of columns, Ari keeps the hierarchy and context that give fields meaning: paths, cardinality, neighboring fields, segment roles, qualifiers, types, constraints, concepts, and code values.
Shape graph matching
A matching is a graph of proposed relations
Ari keeps both structures intact while connecting nodes that express the same role, even when names and nesting differ.
An EDI 204 source shape graph is matched to a canonical shipment target shape graph. EDI 204 corresponds to Shipment using shape context. N1 ship-to corresponds to destination using concept and qualifier evidence. N102 corresponds to destination name. N104 corresponds to destination facility ID using ontology and code evidence. N4 corresponds to destination address by structural role. N403 corresponds to postal code using lexical and type evidence.
Inference proposes edges between compatible source and target nodes. An edge can express direct field equivalence, value transformation, qualifier-dependent meaning, structural projection, or a richer relationship that needs expert review. Each candidate carries confidence, evidence, and traceable rationale instead of appearing as a black-box answer.
Statistical Discovery, Semantic Authority
Ari does not ask one model to decide what a mapping means. Retrieval, embeddings, learned scorers, and rerankers make a large search space tractable by finding and prioritizing plausible relations.
Ontology bindings and declared constraints then test those candidates against type, cardinality, hierarchy, mutual exclusion, and domain rules. A global solver selects a coherent set of relations rather than a collection of locally attractive guesses.
Ari inference pipeline
Learned evidence and explicit semantics work as one system
Statistical and semantic reasoning stay distinct, but they operate over the same evolving relation graph and meet at every consequential decision.
Source and target shape graphs enter two interdependent inference lanes. Statistical discovery uses lexical and bi-encoder retrieval, GBDT and MLP scoring, and cross-encoder, CRF, and graph neural network reranking. Semantic governance uses ontology candidate expansion, type, code, and cardinality compatibility checks, and global hierarchy and exclusion constraints. The lanes repeatedly exchange a shared candidate graph and scored evidence graph before constrained MAP and ILP selection produces a reviewable relation graph with confidence, evidence, and rationale.
Statistical methods discover what might fit. Explicit semantics, constraints, and review determine what becomes governed knowledge. This is Ari's product-level expression of the broader Cohesive principle that statistics discover while semantics commit.
See the inference layers and data flow in Ari Architecture.
Human-Guided Learning
Reviewers accept, reject, or edit candidate relations while refining concepts, crosswalks, ontology bindings, and constraints. Ari curates that feedback before promoting either model updates or governed semantic assets.
Human-guided learning
Review creates two kinds of reusable knowledge
Ari captures corrections as structured feedback, then routes them through controlled statistical and semantic learning paths.
An Ari inference run produces ranked relation candidates for human review. Reviewers accept, reject, or edit candidates and add rationale. Ari curates and evaluates that feedback before routing it into two learning paths. The statistical path creates gold mappings, hard negatives, and preference examples, then trains and validates model updates. The semantic path creates governed relations, concepts, and code crosswalks, then versions and publishes semantic assets. Only evaluated model updates and governed asset versions feed later inference runs.
Accepted mappings become reusable evidence. Rejections become hard negatives. Updated semantics improve candidate generation and compatibility checks. The next inference run starts from a stronger base without silently treating every interaction as truth.
See the complete review workflow in the Schema Matching Studio.
Use Cases and Outcomes
Canonical model adoption
Align legacy schemas and product-specific contracts to shared domain concepts without rebuilding the same decisions for every system.
API and event alignment
Compare evolving service contracts, event payloads, and data products through one governed relation model.
Across these programs, Ari helps teams shorten onboarding cycles, reduce repetitive field-by-field work, improve traceability, and make schema change easier to absorb. Confirmed relations can drive downstream transformation, validation, documentation, and automation.
Domain experts remain the authority. Ari amplifies their judgment by turning mapping decisions into inspectable, testable, and reusable assets instead of leaving them in scripts, spreadsheets, or institutional memory.
Explore Ari
Ari is developed alongside the Cohesive semantic toolchain. Its governed relation model comes from Cohesive.Relations; Ari adds automated inference, semantic enrichment, review workflows, and learning from curated feedback.
Schema Matching Studio
Follow the workflow from imported specifications to reviewed, reusable relations.
ARI Architecture
Examine the inference layers, training data, semantic shape graph, and review loop.
ARI Formalization
Review the formal model for shapes, evidence, ranking, constraints, and globally consistent proposals.