I’m a data science researcher. I help turn messy, real-world data into decisions you can trust, with machine learning and AI that hold up under real constraints.
Whatever you’re working on, let’s talk.
Sound familiar?
- A result that looks great in a notebook, and you’re not sure it survives contact with reality.
- Real-world data that’s messy, incomplete, and never like the demo.
- An AI idea worth trying, but the data is sensitive or regulated.
- A number everyone repeats that no one has actually validated.
- The cost or the risk of running AI creeping up on you.
- A question you suspect the data could answer, if only it were framed the right way.
If any of these feel like your week, you’re in the right place. The messier the question, the more it interests me, and what I care about is giving you an answer you can trust. Here’s how we can work together.
Selected work
Some of what I’ve been working on recently, including code you can open and read.
Automating clinical data entry, fully on-prem
Open source · Python, local LLMs, OCR, REDCap
Manual entry of clinical PDFs into REDCap (a medical-research database) is slow and error-prone. This pipeline reads the PDFs by OCR or embedded text, extracts the fields with local LLMs, validates them against the live REDCap data dictionary, and resolves patient records without duplicating visits, with a human check before anything is imported. Everything runs locally, so the documents and the API token never leave the machine. A study-agnostic engine keeps the sensitive, project-specific parts private.
Keeping sensitive data in-house with local LLMs
Medicilio · Technology advisor · Jul 2025 to Dec 2025 · Milan
The problem: cut administrative workload without sending sensitive data to outside providers. The approach: internal documentation workflows built on local, self-hosted LLMs, so the data never leaves the building and exposure to third-party services stays low.
Making AI’s energy cost measurable
Broken Pot · Data consultant · Dec 2025 to May 2026 · Padua
The problem: much of AI’s cost sits in the energy and infrastructure needed to run it. The goal: turn energy efficiency from a broad ambition into measurable cuts in consumption. My part sits on the data side, making the savings visible and defensible.
Research
I stay active in peer-reviewed research, mostly where clinical questions meet data and machine learning. A selection below, with the full list on Google Scholar.
Let’s talk
Whatever field you’re in, if you’re solving something interesting or just want to think out loud about a problem, I’d like to hear from you. Tell me what you’re working on.
Reach me on LinkedIn Read the blog
