I build analytical systems that make changing data easier to trust, understand and use for decisions.
Data modelling · Reliable pipelines · Data quality and reconciliation · Analytical products
My work sits between the source data and the decision someone needs to make. I think about what one row represents, how a record should be identified, what happens when a load runs twice, and how the final model can be checked against its source.
That means I focus on:
- defining grain, business keys and source contracts before building transformations;
- preserving raw arrivals while creating clean current and historical views;
- making replay, duplicates, late data and corrections explicit;
- reconciling outputs so a dashboard or analytical product can be trusted; and
- documenting the decisions and limits in plain language.
flowchart LR
A["Understand<br/>question · grain · source contract"] --> B["Preserve<br/>raw data · lineage · history"]
B --> C["Trust<br/>quality rules · replay · reconciliation"]
C --> D["Model<br/>facts · dimensions · analytical layers"]
D --> E["Use<br/>dashboards · data products · APIs"]
| Project | Purpose | Current state |
|---|---|---|
| CareerSignal | A Germany-first job and skills intelligence product built around historical vacancy snapshots, skill normalisation and evidence-based career decisions | Charter and Bronze design — private |
| MarketTime | A point-in-time market data platform designed to preserve what was known when, including revisions and late arrivals | Next flagship — local foundation in progress |
These projects are published only when their data-source terms, validation evidence, documentation and ownership walkthrough are complete.
| Repository | What it demonstrates | Classification |
|---|---|---|
| Databricks Write Patterns | The difference between transport and write behaviour, including append, file-level idempotency, duplicate identity and Delta-table history | Technical learning lab |
| SQL Warehouse Course Lab | Source integration, dimensional modelling, referential checks and analytics-ready warehouse views | Guided implementation |
The earlier CV-matching prototype has been archived. Its useful ideas will be reconsidered only inside CareerSignal after the underlying analytical model is trusted.
Every featured project must make six things easy to inspect:
- the problem and intended user;
- the current implementation—not only the roadmap;
- the data contract and architecture;
- validation, replay and reconciliation behaviour;
- decisions, limitations and attribution; and
- the exact boundary between personal work, guided learning and AI assistance.
Employer data, confidential architecture, credentials and unsupported outcomes are never published here.