I am a Computer Science PhD candidate at the University of Milano-Bicocca. My work brings together semantic data engineering, knowledge graphs, agentic AI, workflow automation, research software, and systematic evaluation.
From intent to evidence: natural-language requirements → validated specifications → executable workflows → evaluation and recovery
I develop Python systems that transform heterogeneous data and natural-language requirements into structured, validated, and reusable software and data products.
| 🧠 Applied AI | 🕸️ Semantic data | ⚙️ Reliable workflows |
|---|---|---|
| Agentic systems, structured generation, tool use | Knowledge graphs, entity linking, semantic enrichment | Specifications, code generation, evaluation, recovery |
- Semantic data and knowledge graphs: entity linking, schema alignment, metadata enrichment, RDF, OWL, SPARQL, Neo4j, Cypher, and semantic search.
- Applied and agentic AI: LangGraph, LLM APIs, structured generation, multi-agent orchestration, tool use, validation, and bounded repair.
- Workflow and data engineering: interoperable pipeline specifications, code generation, FastAPI services, PostgreSQL, Kafka, Spark, Airflow, and containerized execution.
- AI evaluation and reliability: execution-based benchmarks, contract validation, failure analysis, confidence handling, provenance, reproducibility, and human review.
| Project | What it demonstrates |
|---|---|
| inLUMEN Code Generation Service | Authenticated FastAPI service that generates and validates executable pipeline components, including Python code, dependencies, Dockerfiles, runtime manifests, and validation reports. |
| Entity Linking Agent | LangGraph-based multi-agent system for contextual entity linking across multiple knowledge sources, with confidence scoring, execution monitoring, and asynchronous APIs. |
| SemTUI | Semantic enrichment platform for transforming heterogeneous tables into structured, knowledge-graph-aligned data products. |
| Notebook Provenance | Static analysis of computational notebooks using AST parsing, data-flow graphs, semantic classification, explainability, and column-level lineage. |
| PipeSpec + OrchSpec | Structured, orchestrator-independent pipeline specifications with deterministic validation and compilation to multiple workflow platforms. |
| Real-Time Fraud Detection | Forked reference implementation of a streaming ML pipeline using Kafka, Spark Streaming, Airflow, MLflow, MinIO, PostgreSQL, Redis, and Docker Compose. |
Represent ──► Generate ──► Validate ──► Execute ──► Evaluate ──► Recover
data code contracts workflows evidence safely
- Reliable workflow generation: moving beyond syntactically plausible code toward contract-aware generation, deterministic validation, execution, and governed recovery.
- Semantic enrichment: connecting heterogeneous scientific and urban data to shared concepts, identifiers, metadata, and graph structures.
- Provenance and reproducibility: extracting lineage from computational notebooks and producing auditable data and software artifacts.
- Sustainable AI-assisted pipelines: evaluating generated workflows for runtime behaviour, quality, and energy and carbon impact through PISCES.
Languages: Python, SQL, Bash
Applied AI and ML: LangGraph, LLM APIs, Hugging Face, scikit-learn, embeddings, semantic search
Data and graph systems: PostgreSQL, PostGIS, Elasticsearch, Neo4j, RDF, OWL, SPARQL, Wikidata
Data and software engineering: FastAPI, Flask, Docker, Kafka, Spark, MLflow, MinIO, Redis
Workflow orchestration: Airflow, Prefect, Dagster, Argo Workflows
Interfaces: REST APIs, OpenAPI, Streamlit, Gradio
Selected machine-learning and analytics projects
- Sepsis Classification with FastAPI — Imbalanced classification, model comparison, hyperparameter tuning, F1/AUC evaluation, and deployment through FastAPI, Streamlit, Docker, and Hugging Face Spaces.
- Expresso Telco Churn Classification — Customer-churn modelling with data preparation, feature engineering, classification, model evaluation, and business-oriented retention insights.
- Indian Startup Data Analysis — Exploratory analysis, statistical hypothesis testing, visualization, and interpretation of investment patterns across industries, locations, and funding stages.
My work spans scientific data infrastructures, smart grids, digital substations, urban and environmental data, health data, entity resolution, and AI-assisted workflow engineering.
Before my PhD, I worked on IEC 61850 digital-substation interoperability at KEPCO Research Institute and smart-metering infrastructure at Landis+Gyr.
I build reusable software, documented APIs, evaluation frameworks, technical demonstrations, and reproducible experiments through international research and engineering collaborations.
Selected publications and research outputs are available through my ORCID.
Technical writing
I write about applied machine learning, semantic data systems, AI-assisted workflows, and practical deployment on Medium.
I am open to applied AI, semantic data, knowledge-graph, data-engineering, AI evaluation, and technical solutions roles in Europe.
