Axiom Policy Reasoner
Engine ready

Policy workshop

Turn intent into enforceable policy.

Use AI to draft. Use deterministic semantics to decide.

1Describe
2Generate
3Validate
4Reason
Demo scenario

Shared semantic layer

Customer Ontology

One schema powers the interface, LLM prompt, validator, and reasoner.

AS
Acme Systems Inc.Customer #1001 · unrestricted
01

Business intent

Describe the new policy

Human authored

Ontology and DSL grammar will be included automatically

Production directionHow this deterministic demo scales Not used in the current runtime

Demo today → production evolution

Keep the trust model. Scale each responsibility.

The LLM proposes. Deterministic systems validate and analyze. A human approves what becomes active.

01

Semantic model & validation

Apache Jena + SHACL

Formalize the ontology as RDF and validate types, domains, required properties, and relationships with a W3C standard.

Today JS ontology + custom checks → Scale RDF/OWL + SHACL shapes
02

Executable decisions

DMN + Kogito/Drools

Publish approved policy as auditable decision tables with typed FEEL expressions, versioned deployment, and deterministic execution.

Today Bounded DSL + AST → Scale Standard DMN runtime
03

Formal conflict analysis

Z3 constraint solver

Prove scope overlap and satisfiability across richer conditions, then return a concrete example that demonstrates each conflict.

Today Interval comparison → Scale Constraint solving

Demo artifacts

What each production tool produces

Representative artifacts for the ADP policy shown in this demo.

Jena + SHACL

RDF policy and validation shape

Jena stores the policy as linked domain facts. SHACL makes its required property, datatype, and unit constraints executable.

policy.ttl 
@prefix ax: <https://axiom.example/policy#> .

ax:ADP_MAX_30 a ax:CreditPolicy ;
  ax:appliesTo ax:Customer ;
  ax:constrains ax:adp_days ;
  ax:maxExclusive 30 ;
  ax:unit ax:DAYS .

ax:CreditPolicyShape a sh:NodeShape ;
  sh:targetClass ax:CreditPolicy ;
  sh:property [ sh:path ax:maxExclusive ;
                sh:datatype xsd:integer ;
                sh:minCount 1 ] .
DMN + Kogito/Drools

Executable decision table

Policy owners review a standard decision table. The runtime records the matching row and returns a typed decision.

customer-credit-policy.dmn
DecisionMaximum ADP DaysU · Unique hit
#Restricted statusCurrent balanceMaximum ADP
1"N"> 100,000 USD< 30 DAYS
2"Y"—Manual review
Z3 constraint solver

Conflict proof and witness

Z3 proves that the active and proposed constraints overlap, then supplies a concrete customer state that demonstrates the conflict.

analysis.smt2 
(declare-const adp_days Int)
(assert (<= adp_days 45))    ; proposal allows
(assert (not (< adp_days 30))) ; violates active
(assert (= adp_days 35))      ; sample customer
(check-sat)                   ; sat → conflict exists
CONFLICT WITNESSrestricted_status = "N"current_balance = 125,000 USDadp_days = 35 DAYS

Incremental adoption

A phased path—not a platform rewrite

  1. ProductionizeVersioned policy API, approvals, audit trail, and shadow testing
  2. StandardizeExecute approved decisions as DMN with regression tests and rollback
  3. FormalizeAdd Jena and SHACL when taxonomy and semantic relationships require it
  4. ExpandUse Z3 for complex overlap, exceptions, dates, and conflict witnesses

Clear boundary: Jena and SHACL validate semantic meaning, DMN executes approved decisions, and Z3 proves constraint conflicts. The LLM is never the policy authority.

Copied to clipboard

Ontology property