Explore. Prototype. Build.

Three recurring phases in the work I do.

1. 01Explore

   Evaluating new technology

   Establishing what a technology can and cannot do, its operational limits, and the cases where it is not the right tool.

2. 02Prototype

   Prototypes and proofs of concept

   Working versions built in short cycles, covering the critical path in full and leaving the rest out. Complete enough to support a decision on whether to continue.

3. 03Build

   Systems in production

   Integration into existing processes, with measurement of results and enough documentation for the system to be maintained by others.

• AI Engineering

  RAG architectures, embeddings and vector databases, LLM orchestration and tool integration over MCP, evaluation and observability. Procedures in which the model handles defined steps inside existing workflows, rather than conversational interfaces.

• Data Systems

  Ingestion pipelines, data quality controls, design of proprietary scoring models and of the methodologies used to validate them.

• Prototyping & New Products

  Moving from an idea or a new technology to a usable prototype, and assessing what it would take to turn it into a product: the data required, running costs, maintenance.

• Automation & Digital Transformation

  Analysis of operating processes and development of the tools that support them: system integrations, internal applications, monitoring.

1. 2024 — Present

   Current

   Italy

   AI Engineer — Applied LLM Systems

   SpazioDati — Cerved Group, today ION

   Evaluation and prototyping of LLM-based systems for existing data products.

   • Design and prototyping of RAG architectures: retrieval strategy, evaluation of results, measurement of reliability. Validated prototypes were taken to production by dedicated teams.
   • Development of LLM-assisted procedures integrated into existing workflows, without conversational interfaces.
   • Work on vector databases and embeddings: vector representation of text, indexing strategies and comparison of retrieval algorithms.
   • Introduction of evaluation and observability tooling (Langfuse) to measure the effect of changes to models, prompts and retrieval strategies.

   TechnologiesPython · LangChain · LiteLLM · Langfuse · RAG · Embeddings · Vector databases

2. 2025 — Present

   Current

   Digital Transformation Engineer

   Prevent

   Analysis of operating processes and development of internal tools. Part-time, in parallel with the role above.

   • Development of an Android reporting application for employees, in daily use.
   • Development of the API connectors between the corporate card system and the expense management platform.
   • Development of monitoring platforms for shifts, sites and emergencies, across dozens of active sites.

   TechnologiesPython · REST API · Android · System integration

3. Sep 2021 — 2024

   Italy

   Data Scientist

   SpazioDati — Cerved Group, today ION

   Design and validation of proprietary scoring models on company data.

   • Design and validation of proprietary scores covering approximately 6 million Italian companies, including the definition of the validation methodology.
   • Among them Cyberscore, which assesses a company’s online positioning, and a trademark score measuring trademark protection activity.
   • Development and maintenance of the ingestion pipelines and data quality controls behind them, on volumes in the order of tens of GB.

   TechnologiesPython · pandas · Data pipelines · Scoring models

4. Aug 2019 — Aug 2021

   Sophia-Antipolis, France

   Software Engineer — Data Distribution

   ALTEN, consultant at Amadeus

   Data Distribution framework, across two teams.

   • Development of the data propagation flow between central databases and their cloud instances, including versioning and consistency handling.
   • Work on central and cloud cache layers, with limited capacity and high update rates. Support tooling, optimization and testing.
   • From 2020, following the reorganisation, data analysis for fraud detection.

   TechnologiesJava · Python · Kafka · Couchbase · Docker · OpenShift

5. 2018 — 2019

   Perugia, Italy

   Machine Learning Researcher — Master Thesis

   Dept. of Pharmaceutical Sciences, University of Perugia

   Machine learning models applied to drug discovery for immunotherapies.

   • Development of a Python workflow for selecting molecules with target properties, based on multimodel stacking.
   • Presented at the CDDD 6th Meeting, Rome, March 2019.

   TechnologiesPython · Keras · TensorFlow · scikit-learn

LinkedIn