Project GAINS

Report on objectives, results, and lessons learned related to Task T2.1 System Design

Giordano, A. – Basile, L. – De Giacomo, A. – Haider, B. - Fruncillo, D. - McGann, O. | 2026-01

This report describes in detail the design of the automation and decision support system of the GAINS project. Objectives, design results, and lessons are presented related to agent-based architecture and the main design artifacts (requirements, data model, interaction contracts, and operational flows) that enable the integration of components and the subsequent validation of the proof-of-concept. The system is designed to provide automated decision support along an end-to-end workflow that includes the ingestion and normalization of heterogeneous data, state and trend analysis, generation of standardized deployment models optimized for environmental sustainability, and production of accompanying technical and procedural documentation. The architecture adopts a modular design that separates observability, inference, and control functions and organizes them into three main interoperable agents: (i) Network Monitor, for collection and read-only analysis of telemetry, construction of historical baselines, detection of anomalies, and impact estimates; (ii) Knowledge Manager, for ingestion, indexing, and synthesis of technical-operational knowledge and constraints (vendor/role/version), with blueprint production reusable; (iii) Network Orchestrator, for translation of intent into governed candidate configurations, accompanied by a change plan, rollback, and assessment of the expected impact tag. The Climate component is integrated as a canonical object of the workflow and enables an assessment that can be attributed by device/service/configuration, explicitly distinguishing between direct measurements and model-based/hybrid estimates, with declared assumptions and confidence levels. The design includes additional governance and security safeguards (audit/traceability, human-in-the-loop, consistency controls, and access policies) necessary to ensure operational consistency and risk reduction, maintaining the modularity of the system and the possibility of independent evolution of the individual agents in the subsequent phases of the design.

Table of Contents

1. Introduction
1.1 Purpose and objectives of the system
1.2 Scope and context 
1.3 Agent-based architecture and divide-and-conquer paradigm
1.4 Industrial replicability and design constraints 
2.  System Requirements
2.1 Functional requirements 
2.1.1 Network Monitor
2.1.2 Knowledge Manager
2.1.3 Network Orchestrator
2.1.4 User Interaction and Integration
2.2 Non-functional requirements 
2.2.1 Security and segregation
2.2.2 Auditability, observability, and reproducibility
2.2.3 Performance and Reliability
2.3 Footprint/climate requirements 
2.4 Table of requirements and verification criteria 
2.5 Traceability matrix Requirements/Interfaces/Artifacts and Test cases
3. System architecture 
3.1 Architectural principles 
3.2 Logical view: components and flows
3.3 Component model: runtime components 
3.3.1 GAINS Control Plane
3.3.2 Network Monitor 
3.3.3 Knowledge Manager
3.3.4 Network Orchestrator
3.3.5 Datastore and artifact management
3.4 Interfaces and boundaries of responsibility
3.5 Main architectural workflows 
3.6 Deployment Considerations and Prototype Profile 
4. Components and interfaces
4.1 Contract conventions versioning, references, error model, RBAC
4.2 Northbound Control Plane Interfaces 
4.2.1 API Monitor 
4.2.2 Knowledge API
4.2.3 Orchestration API
4.3 Internal Control Plane interfaces - agents 
4.4 Contracted liability boundaries
4.5 User-Friendly UI Integration Note
5. Detail Model  
5.1 Data Model Principles 
5.2 Canonical entities and minimum fields 
5.3 Knowledge store: proposed classes/objects
5.3.1 DocSource
5.3.2 DocChunk
5.3.3 Policy
5.3.4 Service Definition
5.3.5 Config Snippet
5.3.6 Blueprint 
5.3.7 Config Snapshot Sanitized
5.4 Artifact store: proposed objects and minimum fields 
5.4.1 Orchestration Job
5.4.2 Candidate Configuration Artifact
5.4.3 Change Plan Artifact
5.4.4 Footprint Estimate Artifact
5.4.5 Approval Decision Artifact
5.5 Knowledge store indexing rules 
5.6 Artifact store indexing rules 
6. Sequence diagram of the main flows
6.1 Sequence overview (A–E flows) 
6.2 Sequence A Monitor query
6.3 Sequence B Knowledge query + blueprint
6.4 Sequence C Intent-to-config job
6.5 Sequence D Approval
6.6 Sequence E Ingest
7. Rationale of technological choices and tools
7.1 Selection criteria 
7.2 Selected stack and role in the system
7.3 Consistency of choices per agent
7.4 Considered alternatives and reasons for rejection
7.5 Conclusions
Bibliography

Keywords

Multi-Agent Systems (MAS), Automated Decision Support, Network Orchestration, Intent-Based Networking, Knowledge Management, Climate-Aware Networking, End-to-End Workflow Automation, Human-in-the-Loop Governance, Telemetry Normalization, Operational Risk Mitigation.

Reference: GAINS deliverables are linked to Work Packages and milestones. For additional materials, visit Resources.