Olympia Dartsi

About

In front of Mt Cook (Aoraki), Aoraki / Mount Cook National Park, New Zealand.

I'm a Fellow Postdoctoral Researcher at Lawrence Berkeley National Laboratory, working at A-Lab, an autonomous laboratory for materials discovery. I build systems that let language models guide real experiments, which means most of my work comes back to one question: when can you trust what a model decided? That question is what's pulling me toward AI safety and evaluations. I'm currently building an evaluation framework focused on model calibration and sycophancy, measuring not just whether a model is right, but whether it knows when it isn't.

I started my scientific journey studying physics at the Aristotle University of Thessaloniki, then moved to Annecy, France for a PhD in experimental particle physics at CERN. If you'd like to know more, you can find my PhD thesis here. After that I moved into data science, applying the same core mindset to different problems: combine theory with data to make sense of complex systems.

I revel in the problem-solving side, but I don't want to stop there. Being able to communicate findings and their theoretical underpinnings to audiences of wildly different backgrounds has always been close to my heart.

When I'm not sifting through bits and bytes, you'll often find me running or cycling in the mountains. Apart from a slight obsession with challenges of all kinds, I just love being outdoors. And then there's painting, my preferred form of meditation. Have a look here if you're curious.

"To learn is a natural pleasure, not confined to philosophers, but common to all men." — Aristotle

Publications

• Automating Chemical Reasoning in High-Throughput Phase Identification With a Probabilistic, LLM-Guided Framework
  Advanced Science (Wiley), 2026.  |  Code on GitHub
• Leveraging unlabeled SEM datasets with self-supervised learning for enhanced particle segmentation
  npj Computational Materials, 2025.
• AlabOS: An Open-Source Platform for Autonomous Lab Orchestration
  Digital Discovery (RSC), 2024.
• PhD Thesis — Experimental Particle Physics, ATLAS Experiment
  CERN / Université Grenoble Alpes, 2019.

• Full list on Google Scholar

Projects

Automated Interpretation Framework

Open-source LLM-guided framework for automating chemical reasoning in high-throughput phase identification, developed at LBNL.

Interactive Dashboard for Materials Research

Interactive dashboard to visualize and interpret materials synthesis and characterization data. Built with Python and HoloViz Panel.

Computer Vision Anomaly Detection

Anomaly detection in MNIST digits using unsupervised learning, with model performance evaluated via ROC AUC.

Parkinson's UPDRS Score Forecasting

Predictions of MDS-UPDRS scores, which measure progression in patients with Parkinson's disease.

Time Series Forecasting

Time series forecasting using the ARIMA model.

Multi-Armed Bandit

A multi-armed bandit implementation using the UCB strategy.

Data Imputation

Evaluating different imputation methods for time series datasets to handle missing values.

EDA Cancer Deaths

Exploratory data analysis (EDA) on global death data from 1990 to 2019.

Google Foobar Challenge 2022

Complete solutions for all levels of Google's secret programming challenge.

Portfolio Projects

A summary of projects I worked on during my PhD and in my current role.

EDA Tips in Restaurant Dashboard

Interactive dashboard exploring restaurant tipping data trends.

Healthcare Data Analysis Dashboard

Interactive dashboard for exploring healthcare outcomes data.

• More projects

Outreach

A collection of my outreach lectures and presentations — from scientific conferences to visual explainers and transitions into data science.

MRS Spring Meeting 2025

Data Infrastructure for Automated Labs and Characterization Frameworks.

AI Fundamentals

Introduction to artificial intelligence for general audiences.

Neural Networks 101

Visual introduction to neural networks and deep learning.

Data Visualisation Techniques

Basics and advanced techniques for visualizing scientific data.

From Academia to Data Science

Mindset and tools when transitioning into the data science field.

PhD Thesis Defense

Summary of research contributions during my PhD at CERN.

No More Scrum for Data Scientists

Reflections on how agile methods like Scrum may not fit scientific workflows.

The Ambiguity of Charisma

Exploring how we define and misinterpret charisma in leadership and communication.

Art Gallery

Painting is my preferred form of meditation — here are a few selected works.