Aurant
Working specification / v1
Mongolia does not have a data problem. It has a connection problem.
Aurant joins the records institutions already hold into one governed model of the country, then closes the loop: detect the problem, find the causes, simulate the options, explain the tradeoff, authorize the action, and measure what actually changed.
- Six planes
- Data, semantic, logic, AI, governance, action
- Ten services
- NEXUS to GUARD, adopted one at a time
- Two languages
- Mongolian and English across every surface
01
The closed loop
Observation becomes a model, the model becomes a decision, the decision becomes an action, and the result of the action becomes new observation. That cycle is the product.
OBSERVE WORLD
|
COLLECT DATA
|
VERIFY DATA
|
RESOLVE ENTITIES
|
UPDATE WORLD MODEL
|
DETECT CHANGE
|
IDENTIFY PROBLEM
|
FIND CAUSES
|
PREDICT FUTURE
|
GENERATE OPTIONS
|
SIMULATE OPTIONS
|
COMPARE OUTCOMES
|
EXPLAIN TRADEOFF
|
RECOMMEND ACTION
|
AUTHORIZATION
|
EXECUTE
|
OBSERVE RESULT
|
MEASURE IMPACT
|
LEARN
|
UPDATE MODEL ---+
^ |
+----------+02
Six planes
Each plane has one job and hands a defined object to the next. Nothing skips the governance plane.
01
Data plane
Universal ingestion. Every source, PostgreSQL, Oracle, SAP, Excel, Kafka, APIs, PDFs, GIS, object storage, sensors, becomes a versioned, provenanced Source. Raw is never destroyed.
02
Semantic plane
Entity resolution and the ontology: objects, properties, links, temporal validity and attached evidence. A graph of reality instead of five thousand tables.
03
Logic plane
Rules, statistics, ML and optimization feeding one decision engine: detect, predict, simulate, then explain, recommend, optimize.
04
AI plane
Orchestrated agents with named tools that read only the ontology they are authorized to read. The model reasons over retrieved evidence; it does not invent numbers.
05
Governance plane
Policy engine before every action: user authorized, agent authorized, data accessible, action permitted, human approval required. Otherwise blocked.
06
Action plane
Notify, update, assign, schedule, call an API, run a workflow, into the external system, into the real world, and the result returns to the data plane.
03
Ten services
The platform is divided into ten services with clean boundaries, so an institution can adopt one and grow into the rest.
01 NEXUS
Data connection
Connectors, credentials, refresh policy, source ownership.
02 FORGE
Transformation & quality
Versioned transformations from raw to curated datasets.
03 IDENTITY
Entity resolution
Canonical entities, aliases, match probability with reasons.
04 ATLAS
Ontology & graph
Objects, properties, links, temporal state, domain models.
05 TRACE
Provenance & lineage
Every number traceable back to a source record and timestamp.
06 ORACLE
Analytics, ML, forecasting
Descriptive, predictive and anomaly services over the ontology.
07 SIM
Optimization & simulation
Scenario trees, tradeoff tables, solver-backed options.
08 AGENT
AI orchestration
Intent parsing, bounded context, tools, evidence-bound answers.
09 COMMAND
Actions & workflow
Proposed actions, approvals, execution into external systems.
10 GUARD
Security, governance, audit
Classification, purpose limits, policy checks, full audit record.
SITUATION ROOM
|
+---------------+---------------+
| | |
ANALYST OPERATOR EXECUTIVE
| | |
========+===============+===============+========
AGENT
|
ORACLE + SIM
|
ATLAS
|
IDENTITY + TRACE
|
FORGE
|
NEXUS
|
DATA SOURCES04
Working principles
Nothing newly collected
Aurant reads records the state and its partners already hold, under the mandate that already covers them.
Explainable, not black box
Every score shows the rules and contributions that produced it. Unexplainable output is discarded the first time it is inconvenient.
Evidence before claim
A property value carries its source, timestamp, transformation, author and confidence. AI output never silently becomes truth.
Human authority
Consequential actions are proposals until a named person approves them, and the approval is part of the permanent record.