Snow detection in PV production time series
Deep learning (LSTM)
Detected snow-induced energy loss directly from PV production data, validated against meteorological models. Published at EU PVSEC 2023.
R&D · production
A one-person consultancy specialising in data science for renewable energy, drawing on a background in reinsurance. End-to-end work: problem framing, modelling, deployment. Registered in Slovakia as oo labs s.r.o.
start a conversation →Forecasting, anomaly detection, and quality control for time-series data — the core demand in renewable-energy and risk applications. Built end to end: research, model, deploy.
Custom models for problems off-the-shelf approaches don't fit. Stochastic processes, optimisation, statistical inference, deep learning — the right tool for the actual problem.
Helping data-product teams scope, prioritise, and ship. Drawing on senior R&D leadership in renewable energy and reinsurance.
Recent and selected projects across solar, reinsurance, and machine learning. See the CV for the full archive.
Deep learning (LSTM)
Detected snow-induced energy loss directly from PV production data, validated against meteorological models. Published at EU PVSEC 2023.
Optimisation + classification
Inferred PV system geometry from production data alone, used to verify mounting metadata at scale. Published at IEEE PVSC 2025.
Empirical model leveraging measurement redundancy
Cross-validated parallel solar instruments to flag drift, mis-orientation, and sensor failure earlier than single-instrument methods. Published at IEEE PVSC 2026.
Statistical model
Detected medium-term signal degradation from soiling using only production data, without requiring on-site soiling sensors.
Retrieval-augmented generation (RAG)
Internal LLM-powered chatbot for querying a historical archive of customer solar reports. Production deployment with citation traceability.
Python library + Airflow
Designed and implemented the data-pipeline architecture supporting all of Solargis' automated QC products. CI/CD via GitLab, deployment via Airflow.
Geospatial app + model evaluation
Built an internal geospatial application for spatial visualisation and comparative evaluation of competing vendor catastrophe models for Japan typhoon risk.
Spatial analytics + mapping
Spatial analysis and visualisation of tropical-cyclone Debbie claims data to inform model-vs-actual loss comparisons and inform future treaty pricing.
Computer vision + generative models
Prototype pipeline combining scene depth prediction, object segmentation, GAN-based image generation, and voice synthesis to assist 3D animators.
Mathematical modelling
Distinction-grade MSc dissertation modelling water flow through gridded urban networks. Supervised by Prof. Gavin Esler.