10/16/2026 updated
100 % available
AI / LLM Engineer & Data Scientist - B2B Contractor
Genova, Italy
Only remote
MSc in Computer Science (Data Science), Università degli Studi di Genova, Italy, July 2024About me
AI engineer (MSc Data Science) building LLM applications end to end: RAG, agents, structured output, evaluation and deployment, in Python and FastAPI. Freelancer with my own Italian Partita IVA, fully remote from Italy, CET hours. Background in information security and on-chain data.
AI AgentsRAGLLMOpenAIAzure OpenAIData AnalysisPython (Programming Language)Machine LearningGenerative AIFastAPI
AI engineer (MSc Computer Science, Data Science, University of Genova) working as a freelancer through my own Italian Partita IVA. Fully remote from Italy, CET hours, invoicing clients in any country. English fluent.
What I build
- LLM applications end to end: retrieval (RAG), agent workflows with tool use, structured output with validation, evaluation and cost tracking, deployment.
- Data pipelines: scraping, ingestion, deduplication, scheduled jobs, LLM steps for scoring and summarisation, delivery to users (for example through a Telegram bot).
- Backends and APIs in Python (FastAPI, SQLAlchemy, Pydantic), with a React front end when needed.
Recent projects
- Global Campaign Localisation Copilot: LLM application that extracts a structured campaign spec from a brief, generates copy per market and channel, and runs an automated brand-compliance check on the output. FastAPI, React, Anthropic SDK, Pydantic. Deployed as a public live demo with a per-visitor spend cap.
- JobWatch: unattended job-discovery pipeline on a Linux server. Several sources with fallback between providers, LLM scoring against a structured profile, rule-based filters, SQLite, systemd timers, Telegram delivery.
- GenovaLive: multi-source content aggregation with change detection, LLM summarisation and scheduled delivery, built to avoid duplicate sends and silent data loss.
- Freelance consulting for small businesses: network and routing fixes, a company website, advice on running a live price-feed application, and an IT and application review with prioritised recommendations. Requirements taken directly from non-technical owners.
Background
- 4 years working with on-chain (blockchain) data: real-time ingestion, signal detection and anomaly detection on noisy public data.
- BSc in Information Technology Engineering (Information Security) and network engineering training (routing, switching, firewall, VPN). Security and data protection are part of how I design systems.
- Earlier: sole IT support for a multi-branch logistics company, later also in sales, so I am used to working directly with clients.
Tools
Python, SQL, Bash, Anthropic SDK, OpenAI SDK, Azure OpenAI, MCP (Model Context Protocol), local models (Ollama, LM Studio), Claude Code, FastAPI, Pydantic, SQLAlchemy, React, pandas, NumPy, scikit-learn, SQLite, MySQL, Docker, Linux, Git, AWS (EC2), Google Cloud, Power BI, Tableau.
Experience
Professional experience0 - 5 years
Freelancer since2024
Languages
EnglishFluentItalianGood
Project history
LLM application for multi-market marketing automation: extracts a structured campaign spec from a brief and brand guidelines, generates channel- and market-specific localised copy, and runs an automated brand-compliance check over the output. FastAPI and SQLAlchemy backend with a direct Anthropic SDK integration, Pydantic-validated schemas and a mock mode for offline development; React front end. Deployed and publicly reachable as a live demo, with a per-visitor rate limit capping API spend. Ships with an interactive masterclass in English and Italian.
Daily job-discovery pipeline running unattended on a Linux server. Several discovery lanes feed one funnel: a tiered fetch cascade falls back across providers when a source blocks or rate-limits; an LLM scores each posting against a structured profile and deterministic filters remove roles that fail on work authorisation, seniority, language or prior application. SQLite persistence, Telegram delivery, systemd timers. A second lane samples the wider market against a frozen panel of companies. Python, packaged with uv.
Resolved dual-WAN routing for a commodities price-data business; designed and built a WordPress website end to end; advised on the operation and positioning of a live gold and silver price-feed application; assessed a second client's IT and application estate for integration and performance issues, delivering prioritised recommendations.