Physics PhD · Scientific computing, Bayesian inference & machine learning
Athens, Greece
I develop scientific software and use statistical and machine-learning methods to infer physical properties from data. My experience spans gravitational-wave astronomy, numerical modelling, Bayesian statistics and PyTorch deep learning, alongside Python and full-stack application development.
My doctoral research at the University of Florida focused on gravitational-wave backgrounds: inferring the emission history of their astrophysical sources, calculating detector responses and correlations, and investigating environmental features in LIGO data. Doctoral research continued through August 2024, followed by research collaboration through September 2025.
I developed a scientific Python package that efficiently computes the angular power spectrum of the astrophysical gravitational-wave background, enabling Bayesian inference on the history of its sources. I also derived and implemented analytic detector-correlation calculations, and contributed spectral analysis and parameter-estimation validation to the collaborative pygwb framework.
- JAX-GW: Python/JAX research software for gravitational-wave modelling, detector response and inference.
- PhD thesis: the scientific questions, methods and results.
My current independent research explores neural posterior estimation: training PyTorch conditional-density models, including mixture-density networks, neural spline flows and ensembles, for nonlinear, multimodal and selection-biased simulation problems. I compare learned posteriors with numerical and sampling references, using calibration and posterior-predictive checks to assess their behaviour.
- batch-doc-vqa: schema-constrained document extraction and evaluation of local and hosted language/vision models, with quality, runtime and cost reporting.
- Instructor Pilot: a co-developed Django/React application connecting document processing, human review and feedback, relational data and Canvas APIs.
- Latin-rectangles: combinatorial algorithms and polynomial arithmetic in a Python package and command-line interface.
My main tools are Python, JAX, NumPy/SciPy, PyTorch and scikit-learn, with Django, React, TypeScript and SQL for applications.
My GPU and parallel-computing experience includes CUDA, threading, SIMD and ISPC coursework, and single-node multi-GPU training reproduction with PyTorch DDP/NCCL.
I use Codex in daily engineering work to decompose tasks and direct parallel implementation and review. I assess numerical changes with physical-limit tests and independent reference calculations, and integrate changes after checking their behaviour.