Selected client and independent work across machine learning, analytics and data systems.
Private client · Production project
I designed and built a production-facing analytics application for a private client. It brings product catalog and sales data together so commercial performance can be explored in one place.
Data ingestion and transformation flows, catalog-to-sales matching, analytics logic, interactive filtering and product, customer and order-level views.
The project involved ETL-style processing, validation, backend work, data refresh flows and the full delivery of a usable analytics product rather than a notebook prototype.
I iterated on which questions and metrics were actually useful, keeping the interface focused instead of turning it into a generic BI dashboard.
2026
Implemented a character-level decoder-only Transformer in PyTorch from first principles, including the data pipeline, causal self-attention, multi-head attention, residual blocks, evaluation, checkpointing and autoregressive generation.
View on GitLab2025
Built a probabilistic pipeline for ATP match outcomes and evaluated the signal against bookmaker odds using calibration, Brier score and Kelly-based decision analysis rather than accuracy alone.
View on GitLab2025–now · Livesport
Worked across ETL pipelines, data engineering tasks, ad hoc analyses, process automation with Airflow, and data science work in a production product environment.