Platform

Ten services, six planes, one loop

Aurant is not one application. It is a spine of services with clean boundaries, so an institution can start with connection and lineage and grow into simulation, agents and governed action.

Service catalogue

Each service owns one contract and can be deployed independently.

01 NEXUS

Data connection

Registers every source with its connector, credential reference, schema, sensitivity, refresh policy, owner and provenance. Databases, files, streams, APIs, documents, GIS and sensors enter the same way.

02 FORGE

Transformation & data quality

Raw dataset to version to transformation to curated dataset, with quality checks at each step. The raw layer is immutable, so any curated value can be replayed.

03 IDENTITY

Entity resolution

Decides when records from different sources describe the same person, company or asset, and keeps the uncertainty, with the contribution of each matching signal.

04 ATLAS

Ontology & knowledge graph

The operational model of the country: typed objects, properties, links, temporal validity and the actions each object supports.

05 TRACE

Provenance & lineage

Answers 'where did this number come from' down to the transformation id, raw dataset version, source system and ingestion timestamp.

06 ORACLE

Analytics, ML & forecasting

Descriptive, predictive and anomaly services: SQL, statistics, graph algorithms, time series, geospatial models, forecasting.

07 SIM

Optimization & simulation

Builds scenarios off the current state model, runs solvers, and returns comparable outcomes with their assumptions attached.

08 AGENT

AI orchestration

Parses intent, evaluates permission, retrieves ontology context, calls tools, and only then lets a language model reason, over bounded, cited context.

09 COMMAND

Actions & workflow

Turns an approved decision into a concrete write: notify, update, assign, schedule, call an external API, start a workflow, and records the result.

10 GUARD

Security, governance & audit

Classification and purpose limits inside the object itself, policy evaluation before every read and write, and an immutable audit record of both.

From logic to action

The path a decision takes, and the checkpoint it cannot pass without authorization.

=========== 03. LOGIC PLANE ============================
  Rules      Statistics      ML      Optimization
     |            |          |            |
     +------------+----+-----+------------+
                       |
                DECISION ENGINE
                       |
       +---------------+---------------+
       |               |               |
     detect         predict        simulate
       |               |               |
     explain        recommend       optimize
       +---------------+---------------+
                       |
=========== 04. AI PLANE ===============================
                AI ORCHESTRATOR
                       |
       +---------------+---------------+
       |               |               |
      LLM        Agent planner   Specialist models
       +---------------+---------------+
                       |
                  TOOL ENGINE
                       |
          reads authorized ontology only
                       |
               PROPOSED DECISION
                       |
=========== 05. GOVERNANCE PLANE =======================
                 POLICY ENGINE
        user? agent? data? action? human?
                       |
              +--------+--------+
             NO                YES
              |                 |
            BLOCK          ACTION ENGINE
                       |
=========== 06. ACTION PLANE ===========================
  notify | update | assign | schedule | API | workflow
                       |
                EXTERNAL SYSTEM
                       |
                   REAL WORLD
                       |
                   NEW RESULT ------> DATA PLANE

Reference technology

Nothing here is proprietary. The advantage is the ontology, the governance and the operating loop, not the engines underneath.

Ingestion

Streams and connectors

Kafka, Debezium, Airbyte and purpose-built connectors for ministry and enterprise systems.

Storage

Object store and databases

S3-compatible object storage for raw, PostgreSQL for operational state, ClickHouse for analytical scale.

Transform

Batch and stream compute

Spark, Flink, Polars and dbt, with lineage emitted from every job.

Index

Search, graph, geo, vector

OpenSearch for text, a graph layer for links, PostGIS for geography, pgvector for retrieval.

Reason

ML and optimization

Python, PyTorch, XGBoost and MLflow for models; OR-Tools and Pyomo for solvers.

Orchestrate

Durable workflow

Temporal for long-running, resumable action workflows with human approval steps.

Authorize

Policy

OPA plus an attribute-based access layer expressed against ontology objects, not tables.

Observe

Telemetry

OpenTelemetry, Prometheus and Grafana across services, models and agent runs.