Agentic DPP Platform · v3.0 · 7 Services · 8 Layers · Click any node to explore

DPP Agentic Microservice Stack

Agent-augmented microservice architecture for Digital Product Passport generation. Each microservice is paired with a purpose-tuned AI agent via MCP interfaces, coordinated by an Orchestrator Agent — operating within a Trust & Governance Layer enforcing policy-aware, deontic behavioural boundaries.

v3.1 · Agentic · 8 Layers · 9 Services
Human
Trust
Orchestrator
Services
Publishing
Analytics
Impact
Infra
Agent-augmented service
MCP interface
Layer 01
Human
Interaction
Users, domain experts, final-review gate
🏭 DPP Creator (Manufacturer)
🧪 Domain Expert / Reviewer
🏛 Regulator
♻ Recycler / EoL Operator
👤 Consumer
👁Human-in-the-loop review gate — final approval before DPP publication
User intent · document pool upload · review/approve triggers · access queries
Layer 02
Trust &
Governance
Policy-aware · deontic constraints · hallucination prevention · reliability
🛡
Policy Awareness Engine
Deontic logic · OWL/SHACL · Prohibitions / Permissions / Duties
Defines validity space — behavioural boundaries within which all agents must operate. Prevents out-of-scope LLM actions.
🔍
Reliability & Hallucination Guard
Neurosymbolic AI · RAG grounding · confidence thresholds
Validates agent outputs against knowledge graph + RAG retrieval. Flags low-confidence or ungrounded assertions before they propagate.
⚖
Compliance Validation Agent
LLM + SHACL · ESPR Art. 7 · field-level audit
Independently checks whether extracted & transformed DPP data actually satisfies applicable regulatory fields. Surfaces non-conformities.
📚
RAG Knowledge Base
Document index · regulation corpus · ontologies
All agents query here. EUR-Lex, ESPR delegated acts, product standards, supplier datasheets — vectorised and indexed.
Policy constraints → Orchestrator · validated agent actions only
Layer 03
Orchestrator
Agent
Master agent — plans, delegates, tracks pipeline execution
🤖
DPP Orchestrator Agent
LLM · LangGraph / CrewAI · task planning · sub-agent delegation
⚡ Master Agent
MCP Server
Receives
User intent + product metadata · Policy constraints from Trust Layer · Tool call results from sub-agents
Plans & Delegates
Decomposes DPP task into sub-tasks → routes to Legal, ETL, Transform, Load, Publish, Analytics agents via MCP
Controls
Pipeline sequencing · retry logic · human escalation on low confidence · final DPP assembly
MCP tool calls → service agents · Kafka events · async results returned to Orchestrator
Layer 04
DPP Pipeline
Services + Agents
ETL microservices, each paired with a purpose-tuned agent via MCP
→
⇄
Data Extraction & Mapping
Apache Camel · OCR · RAG
🤖 Extraction Agent
MCP
02
→
⟳
Data Transformation Service
JSONata · XSLT · Schema Registry
🤖 Transform Agent
MCP
03
→
⬆
Data Load Service
Spark · PostgreSQL · S3
🤖 Load & Validate Agent
MCP
04
Validated canonical DPP → publishing targets · registry
Layer 05
DPP Publishing
& Registry
Multi-spec emission · GS1 Digital Link · lookup
📡
DPP Publishing Service
EU ESPR · AAS SDK · UNTP · W3C VC · GS1
🇪🇺 EU DPP 🔷 AAS DPP 🌐 UNTP DPP
🤖 Publishing Agent
MCP
05
▦
DPP Registry Lookup Service
Elasticsearch · GS1 Digital Link · OPA ABAC
🤖 Lookup Agent
MCP
06
Published DPPs → quality scoring · completeness · TCO modelling · financial impact assessment
Layer 06
Analytics
& Impact
Quality · TCO · financial impact · CSRD bridge
◈
DPP Analytics & Quality Service
Apache Flink · Great Expectations · Grafana · OpenLineage
Mode A · Single DPP
Completeness · consistency · gap detection · SVHC · compliance readiness
Mode B · Portfolio
Cross-DPP benchmarking · anomaly detection · trend heatmaps
🤖 Analytics Agent
MCP
07
💰
TCO Explorer
ISO 15663 · LCC model · Monte Carlo · cost ontology
Total Cost of Ownership modelling across full product lifecycle — acquisition, operation, maintenance, disposal — using DPP data as input.
🤖 TCO Agent
MCP
08
⚖
CSRD ESRS Double Materiality
Financial Impact Estimator
ESRS E1–S4 · EFRAG DM · IRO model · scenario analysis
Inside-Out 🌍
Impact on environment & society from product operations — GHG, material toxicity, labour
Outside-In 📈
Financial risks & opportunities from sustainability matters — regulatory exposure, market shifts
🌉 Bridge to CSRD / ESRS sustainability reporting — IRO assessments pre-populated from DPP data
🤖 Impact Agent
MCP
09
Layer 07
Infrastructure
& Data
Persistence · messaging · observability · secrets
🗄PostgreSQL / TimescaleDB
⚡Apache Kafka
📦S3 / MinIO Object Store
🔍Elasticsearch
📊Prometheus + Grafana
🔐HashiCorp Vault
📋Confluent Schema Registry
☸Kubernetes + Helm
🔌MCP Tool Registry
Agentic Architecture Principles
⚡
One Orchestrator Agent decomposes the DPP task and delegates to specialised sub-agents via MCP tool calls — classic orchestration pattern
⚡
Each microservice exposes an MCP server interface — agents interact with services without bespoke integration code per pair
⚡
RAG framework indexes all document pools, regulation corpora and ontologies — agents query it for grounded, hallucination-free responses
Trust & Governance Layer
🛡
Policy Awareness Engine defines behavioural boundaries using deontic logic (Permissions / Prohibitions / Duties) — constrains what agents may do
🛡
Neurosymbolic AI approach: knowledge graphs + LLMs — symbolic component enforces hard constraints, LLM handles flexible extraction
🛡
Human-in-the-loop final gate — user reviews assembled DPP before publication; escalation path for low-confidence agent outputs
Research Contribution (CORNET)
R
Novel framework: deterministic microservice backbone + agentic AI layer — microservices provide process certainty, agents provide adaptive intelligence
R
Compliant Agents — design and evaluation of policy-aware, legally/ethically bounded LLM agents for regulatory compliance tasks
💰
TCO Explorer applies ISO 15663 LCC modelling to DPP lifecycle data — enabling cost-transparent product decisions across acquisition, operation and disposal
⚖
Financial Impact Estimator bridges DPP data to CSRD/ESRS reporting — double materiality IRO assessments pre-populated from product passport data fields