samuel@portfolio:~/portfolio % cat intro.md
London · Data Engineer · MSc Business Analytics, UCL
Most of the work here is about taking messy source data and making it usable. I am currently completing an MSc in Business Analytics at UCL in London.
Interested in data engineering, analytics engineering, and applied machine learning infrastructure.
dataset metrics
| 1,216 | games in flagship dataset |
| 8 | sources joined into pipeline |
| 62 | features in final output |
samuel@portfolio:~/work % python pipeline.py --info
featured work
Built around 1,216 indie games and eight data sources, this project moves from collection and scraping through Spark processing into a dataset that is actually usable for analysis.
8 integrated data sources · PostgreSQL + MongoDB storage · Apache Spark processing · Parquet + DuckDB analytics · Success tier classification
>open projectsamuel@portfolio:~/projects % ls -la
The project focuses on disciplined evaluation rather than inflated forecasting claims, using walk-forward validation, calibration, and model-odds blending to test whether the pipeline adds signal beyond the market.
Bet365 and Pinnacle benchmarks · Best single model: 0.9678 log-loss · Best blend beats Bet365, p=0.007 · Dixon-Coles season simulation · 6.7/20 exact positions
>open project-02Built to a UCL Data Engineering brief requiring RAG, graph databases, LLM agents, data lineage, and API deployment — all against live, real-world news pipelines.
PostgreSQL · MongoDB · Neo4j · ChromaDB · LangChain agent with 6 tools · Airflow orchestration · OpenLineage / Marquez lineage · FastAPI + MCP interface · Docker Compose (7 services) · spaCy NER · sentence-transformers
>open project-03A supporting analytics project focused on the less glamorous but more useful work of cleaning structure, shaping features, and producing interpretable segments.
Analytics workflow · Feature framing · Decision-ready outputs
>open project-04samuel@portfolio:~/notes % vim notes.md
What broke, what became clearer, and what changed once the work had to run repeatedly rather than just succeed once in a notebook.
Some thoughts on traceable movement, storage boundaries, and why quiet documentation habits matter more than people admit.
