Portfolio – projects in data science, data architecture and AI

1

Control over complexity

Architecture and transparency as the basis for good decisions.

Question: Sectors work with reference architectures, such as CORA and VERA for housing associations and NBility in the energy sector. These often remain on paper. The organisation's own systems, processes and data do not follow them, and overview fades as the landscape grows.
Approach: For housing associations and for a grid operator, I translated the reference architecture into the organisation's own system, process and data architecture. Then I took it off paper. I organised systems and processes so that they demonstrably comply, and a change and governance process keeps them aligned. With data architecture as code, the architecture became a living document that shows in real time how it is implemented and where it deviates from the standard.
Result: An architecture that is not only described but also works. Recognisable in systems and processes, lastingly assured and always up to date.
Lesson: A reference architecture only adds value when it steers daily practice. Architecture is a steering instrument, not documentation after the fact.

Question: Asset data flowed from the central GIS system to ERP, maintenance planning and data warehouses. All logic sat undocumented in code, so the impact of a change could not be overseen.
Approach: I developed a data lineage solution that makes the data flow visible at column level. I also delivered a Python application that converts the data warehouse into a data model as code, automatically discovers relationships between data and tests their quality.
Result: Control over the impact of changes, consumers informed in time and documentation that is always correct.
Lesson: How data really flows through systems, you only see in the technology itself. Transparency starts with reality as it was built.

Question: The facilities department of an academic hospital wanted to professionalise its information provision. An overview of the existing data architecture was missing.
Approach: I mapped the current architecture and developed a roadmap for step-by-step professionalisation. During delivery I led junior consultants, who developed the data warehouse and dashboards.
Result: A supported course from analysis to delivery, in clear steps.
Lesson: A roadmap only works when it connects to where an organisation stands today.

2

Predicting and automating where it demonstrably pays off, designed to be verifiable.

Question: Budgets and plans often rely on last year's figures. How many homes will become vacant, how long will they stay empty, and what will maintenance cost in the coming years?
Approach: Using classic machine learning and forecasting methods, I predict among other things tenancy turnover, vacancy duration and maintenance budgets. The models also reveal the trends and drivers behind the costs, making clear where a development comes from.
Result: More control over budgets, better portfolio management and a well-founded financial plan.
Lesson: A good forecast shows not only what will happen, but also why. That insight is what makes it possible to steer.

Question: Analysts receive questions that require deep knowledge of both the subject matter and the underlying data. An AI agent can help, but only if its answers can be trusted.
Approach: I developed an AI agent that supports analysts in writing queries and answering data questions. The emphasis was on the preconditions. I established the right guardrails and built a ground truth dataset to test the answers against. I also made the agent's reasoning transparent, so each answer shows how the figures were arrived at.
Result: A proof of concept that showed what AI delivers here, and which preconditions are needed to use it responsibly.
Lesson: An AI application is only usable when its outcome is verifiable. Validation belongs in the design, not afterwards.

3

Data platforms and management information that make people's work easier.

Question: Housing associations have similar management questions. Yet each kept building its management information from scratch.
Approach: I developed a data platform as a product, from market proposition and pricing strategy to architecture. The platform consists of a load process, a data warehouse and standard dashboards across all data domains, based on the sector standards CORA and VERA. As lead architect I led the development team and delivered the complex parts myself.
Result: A reusable platform that gives housing associations reliable management information quickly.
Lesson: A foundation based on sector standards makes a solution scalable and prevents architectural debt.

Question: Controllers spent a lot of time collecting and cleaning data in Excel. Little time was left for analysis and advice.
Approach: I improved the architecture and quality of the data warehouse. Per department we defined the key indicators together with staff. I automated the monthly and quarterly reports based on current data.
Result: A complete management report. Controllers could focus on their profession again.
Lesson: Automation is not a goal in itself. The gain lies in the attention freed up for the work that matters.

A data strategy co-created with a housing association.
Advice and training on adopting process mining.
Insight into supply-chain collaboration with contractors, based on internal and external data.
Management reports and Power BI training.
A maturity model that helps housing associations become data-driven step by step.

4

Data quality and knowledge transfer as a fixed part of the work.

Question: A housing association is accountable on many subjects. That requires data you can trust.
Approach: I delivered a data quality dashboard that tests each criterion against the standard the association sets itself. I also advised on assurance and governance following DAMA-DMBOK.
Result: Data quality became a continuous and measurable process, instead of a periodic clean-up.
Lesson: You improve data quality with ownership and a fixed place in operations.

Question: A housing association wanted to digitise its management information, aiming for fewer errors and faster course correction.
Approach: I designed and delivered a new data warehouse with dashboards per department. I trained two IT staff members during design and delivery. Together we deliberately chose a manageable solution.
Result: After completion the association took the solution fully into its own management.
Lesson: A solution only succeeds when the organisation can carry it itself. Knowledge transfer starts on day one.

Question: The controllers of a rehabilitation centre prepared their reports in Excel. They wanted to move to professional, reliable management information.
Approach: I developed a multi-day Power BI course and used it to train the control team. I also delivered management reports and advised on the data architecture, set out in a roadmap.
Result: A control team that works independently with Power BI, and a clear course for further development.
Lesson: Teaching and inspiring people makes change last.