Applied AI Engineer & Solution Architect | Ontology Engineering, Knowledge Graphs and Governed LLM Agents | AI Governance
Psychology-informed governance · Evidence-led evaluation

I design and build AI systems where decisions must stay explainable, testable and accountable: LLM agents, retrieval-augmented generation (RAG), knowledge graphs and ontologies, with human oversight built in from the start. My core technical focus is ontology engineering: I model domains as ontologies and knowledge graphs, and I build AI agents that use that data only through typed, governed queries and tools.

I have almost 20 years of experience, 18 of them at Enel Group, and I have led teams since 2008. From June to September 2026 I was Chief Technology Officer and Head of Research and Innovation at Sistemi & Automazione. Today I run Neuromorphic Inference Lab, my independent open-source initiative, and I am studying for an MSc in Artificial Intelligence at the University of Liverpool.

I am looking for Applied AI Engineer, AI Solution Architect, knowledge-graph and AI deployment roles.
Based in Rome, Italy · Italian (EU) citizen · available immediately · open to relocation within the EU

• Machine learning in production (Enel Group, 2017–2023). In a six-person Agile team, I co-built and maintained for six years a machine learning platform in production on AWS for predictive maintenance and investment targeting, from requirements to production, adoption and ongoing operation. I selected Databricks, personally developed the platform's entire ETL layer (Apache Spark, SQL, Python) and did the modelling: feature selection, XGBoost training and evaluation, and thresholds for operational use. The platform contributed to reducing average customer interruption minutes by up to 30% in some territories (internal before-and-after measure).
• Platforms in production, managed as CTO (2026). I managed two platforms in production: a graph-analysis platform with a multi-agent tool that users could query about the graph data, and a due-diligence platform with about 60 connectors to different data sources. I handled their production issues end to end. Both platforms existed before I joined.
• AI governance (2026). As CTO, I authored the company AI policy (EU AI Act, GDPR and Italian Law 132/2025, prepared with a view to ISO/IEC 42001 certification): risk-based classification of AI use, human oversight and graduated autonomy levels for AI agents. I also authored the guidelines for AI-assisted software development, translating recent academic literature into controls for requirements, testing, traceability and validation.
• Multi-agent architecture (2026, design stage). I defined the architecture of a cloud-native graph-intelligence platform: multi-agent AI over an ontology layer and knowledge graph, agents under human supervision, EU AI Act, GDPR and data-sovereignty requirements, and a verification plan set before development.
• Teams and capability. As CTO, I established the technology function (5 direct and 15 functional reports). At Enel, I created the unit that trains a technical organisation of 1,500 people; at the end of 2025 it ranked first of the 12 training units in Italy for training hours per person.

In development: a runnable implementation on a wholly synthetic scenario, with qualification in progress. OCOR is a sovereign, vendor-neutral semantic core. It connects ontology definitions, evidence, governed decisions and controlled actions, so that software and AI agents work against explicit contracts and authority limits.

• Designed: an ontology-as-code metamodel, with modules and imports pinned by version and digest, typed objects and interfaces, n-ary links with named roles, constraints, bitemporal data, security markings, and contracts for named queries, functions, actions, events and MCP tools.
• Built: the ontology compiler. It type-checks YAML or JSON ontology sources, rejects unresolved or cyclic references, and turns each release into a deterministic, content-addressed canonical model. Every release gets a semantic diff and a compatibility class (identical, compatible, breaking); a breaking release is rejected without a migration plan.
• Built: governed access for agents. They use only typed, policy-filtered named queries and typed tools, with budgets and a kill switch. An end-to-end test crosses all subsystems on synthetic data; checks run in CI with GitHub Actions.

Scope boundary: this is an implementation in development, not a production system. The repository's own records state that qualification is in progress and that runtime conformance and production readiness are not established.

Runnable proof of concept. Graph-RAG over a codebase, with event-driven updates and vector, full-text and graph retrieval. Services in Python, Go and Rust; Neo4j, Qdrant, OpenSearch, Kafka and Debezium, PostgreSQL, Redis.

Scope boundary: this is a runnable proof of concept, not a production service.

Proof of concept with a live demo on synthetic data and one real routed model call. A governed AI workflow that routes the model call to a provider by data-sovereignty rules, requires human approval, records every decision and tracks latency, token use and cost. Next.js, FastAPI, LangGraph, PostgreSQL.

Scope boundary: this is a fixed, typed workflow with governed model invocation. It is not a general-purpose multi-agent runtime, a restart-safe distributed executor or a multi-tenant production platform.

Runnable architecture proof of concept. The core of a policy-governed, AI-assisted software factory: contract-first schemas, an explicit state machine, PostgreSQL leasing, data-residency and cost-aware model routing, and fail-closed controls. Its four-role agent engine (architect, implementer, adversarial reviewer, diagnostician) takes coding agents from specification to tested code.

Scope boundary: this is an inspectable governed core and architecture proof of concept, not a complete production software factory.

Live demos on simulated data. In the onboarding orchestrator, an agent runs an onboarding workflow across simulated systems through tool calls. In the hiring panel and the innovation board, six agents assess a case from different perspectives. In all three, code validates the agents' output and a person decides.

An open-source ML application: fault-risk scoring for medium-voltage lines, with time-based validation, reproducible training and evaluation, MLflow artefacts, Docker packaging and GitHub automation.

Scope boundary: the separate live FastAPI demonstration endpoint is rule-based. This is an independent project, not an Enel system, and not evidence of continuous trained-model serving.

A formal control framework in which an LLM agent judges improvement or deterioration from an external, auditable indicator. Theoretical preprint with public simulation code; not peer reviewed. ResearchGate, DOI 10.13140/RG.2.2.25820.55685

Formal training in Psychological Sciences gives me a complementary, non-anthropomorphic lens on authority bias, conformity, groupthink, social loafing and confidence dominance. I do not assume that LLM agents have human motives. I translate these concepts into observable failure hypotheses, structural controls and evaluation designs.

The designs I have proposed, all testable, cover:

• separating procedural from epistemic authority;
• structured dissent, falsifiable objections, blind review and external verification;
• role permutation, agent ablation and adversarial-leader testing;
• compute-equivalent single-agent baselines;
• measures of task correctness, dissent robustness, cascading error, traceability and cost.

This work currently supports conceptual architecture and evaluation design. Proposed controls and measures stay labelled as proposals until they have been implemented and benchmarked.

• Ontology engineering: ontology-as-code, OWL 2, established methodologies (NeOn, METHONTOLOGY, Ontology 101), competency questions, property-graph data modelling, entity resolution, and the ontology as a governance layer for data, queries and actions.
• Architecture: governed integration, model routing, retrieval, APIs, event-driven data flows and human oversight.
• Agentic governance: roles, permissions, explicit state, structured dissent, external evidence, traceability and fail-closed controls.
• Evaluation: explicit assumptions, success and failure criteria set in advance, telemetry, and validation against observed results.
• Knowledge graphs: Neo4j, Apache Jena Fuseki, TigerGraph, ArangoDB; Cypher and SQL/PGQ; GraphRAG and LLM-based entity and relation extraction.
• Stack: Python, SQL, TypeScript; FastAPI; LangGraph, LangChain, LlamaIndex, Semantic Kernel, MCP; Qdrant, OpenSearch; PostgreSQL, Redis, Kafka and Debezium; Databricks, Apache Spark, MLflow, DVC, Weights & Biases; Docker, Kubernetes, Terraform, AWS, GitHub Actions; Claude and Codex as daily coding agents.

Question → literature review → explicit assumptions → success and failure criteria → instrumentation → validation. I separate evidence, inference and hypothesis, then revise architectural decisions against observed results and documented limitations.

• Chief Technology Officer & Head of Research and Innovation, Sistemi & Automazione S.p.A., Rome — June 2026 – September 2026.
• E-Distribuzione S.p.A. (Enel Group) — May 2008 – June 2026: Head of Training Unit; Senior Project Manager | Digital Transformation Lead (monitored Grid Blue Sky, Enel Grids' global digitalisation programme); Innovation & Agile Ambassador | Senior System Designer; Project Manager.
• MSc Artificial Intelligence, University of Liverpool — in progress, expected August 2027; current average in the Distinction band.
• Postgraduate Certificate in Data Science and Artificial Intelligence, University of Liverpool — academic requirements completed; formal award pending.
• MSc Management and Innovation (LM-77), Universitas Mercatorum.
• BSc Psychological Sciences and Techniques, Universitas Mercatorum.

Neuromorphic Inference Lab is an independent open-source initiative. I define architectures and specifications, direct AI-assisted implementation with coding agents (Claude, Codex), review code and tests, and remain accountable for validation, limitations and deployment decisions. The repositories demonstrate inspectable patterns, bounded proofs of concept and one implementation in development (OCOR). They are not presented as customer production platforms, validated research frameworks or evidence of years operating autonomous multi-agent runtimes at scale.

• LinkedIn
• Website
• ORCID
• Email