I am a Senior Applied Scientist helping organizations turn unstructured data into tested prototypes, decision-grade insights and integrated product features.
For 15+ years I have been in the role technical founder and founding engineer at multiple early-stage startups in Iceland, Netherlands, United Kingdom, and Spain.
I have built and deployed AI/ML systems and data pipelines that have processed millions of data points, providing among other things, age-appropriate search results for children, market insights for real estate investors, and impact information for urban planners.
I have broad background working on applied artificial intelligence (AI), data science, machine learning (ML), natural language processing (NLP) and search technology, and specialization their application to urban planning and sustainability.
Service Offering
Search and RAG
Surface relevant information quickly to support user-facing search and evidence-backed recommendations (focused crawls, retrieval pipelines, ranking evaluation).
Spatial Informatics
Turn fragmented geographic data into actionable insights for planning and sustainability decisions (data integration, spatial joins, mapping, small-scale modelling).
ML & LLMs
Automate information extraction, content classification and decision-support with task-specific models and evaluation benchmarks (fine-tuning, evaluation collections, lightweight distillation).
Principal Clients
Early Stage Startups
Rapid hypothesis validation, demo-ready prototypes and experimental product features.
CITIES & Urban Practitioners
Data-driven insights and case study reviews for planning and policy.
Engagement Models
Data Sprint (2–4 weeks)
Rapid exploratory data analysis to surface insights, opportunities or risks.
Deliverables: Notebooks, stakeholder-ready reports, interactive data stories.
Prototype Sprint (4–8 weeks)
Working data-driven prototype that demonstrates feasibility with customer-facing problems.
Deliverables: Prototype code, demo and documentation.
Product Feature Build (3–6 months)
Production-ready features built alongside client engineers; I focus on hands-on model building, evaluation and integration.
Deliverables: Production models, tests, benchmarking, documentation and handover notes.
Why work with me
- Fast, hands-on execution that de-risks technical hypotheses and produces tangible artefacts for demoing to senior management, customers and investors.
- Domain expertise in urbanism and sustainability for better problem framing.
- Collaborative approach — I work alongside in-house product and engineering teams to move from prototypes to products.
Preferred Stack
I work mostly within the following stack but am flexible to adapt to client needs and setup.
Data Sprints
Python • Jupyter Notebooks
Pandas • GeoPandas • Leaflet • Seaborn
Prototype Sprints
React.js • Flask • FastApi
Product Feature Builds
Python (uv, ruff, pyright, pytest, pre-commit, github actions)
Scikit-Learn • PyTorch • LangChain • Pandas • GeoPandas
PostgreSQL • PostGIS • Elasticsearch • Solr • Nutch
AWS • Azure • Google Cloud • OVH
Past work (selected examples)
Content Understanding and Classification
Automated manual content curation for urban planning knowledge base using LLM-based annotation models that extracted quantitative impact metrics and qualitative insights from unstructured policy documents, research articles, press-releases and blog posts. Built custom evaluation framework ensuring annotation quality matched domain expert standards, enabling Urbanixm to scale its database while staying a one-person startup. (Urbanixm)
Replaced labor-intensive manual processing of financial contracts with fine-tuned LLM-based information extraction system, eliminating error-prone human review bottlenecks. Enabled Banqora to demonstrate AI-powered post-trade processing to early financial institution prospects and investors for their seed-round. (Banqora)
Built in-house Dutch readability classifier with custom evaluation benchmarks that outperformed commercial alternatives, eliminating vendor dependency on core technology. Enabled age-appropriate search for children and content editing tools for educators. (WizeNoze)
Enabled multilingual brand monitoring (EN/ES/CA) without per-language labeling costs through self-supervised learning on Wikipedia/Wikinews data. Language-agnostic approach allowed brandcrumb to expand into new markets without expensive annotation projects (brandcrumb).
Data & Information Processing Infrastructure
Transformed Urbanixm from concept to functioning knowledge base in months rather than years through automated web crawling and content annotation of millions of urban planning documents. The end-to-end pipeline enabled me to demo a working product for seed fundraising. (Urbanixm)
Unified 15+ fragmented real estate data sources into unified API (250+ time-series, 10+ combined indices) that became Placemake.io’s core platform differentiator. Automated integration reduced new data source onboarding time, enabling rapid response to investor information requests and faster customer development. (Placemake.io)
Proofs-of-Concept and MVPs for Early Stage Startups
Shipped interactive impact information MVP in 3 months that became Urbanixm’s primary customer discovery tool, testing value propositions with city planners, engineering firms, and urban think tanks before committing to full build. Flexible visualizations enabled me to iterate product-market fit in real conversations rather than theoretical planning. (Urbanixm)
Delivered working carbon-insetting prototype in weeks that validated technical feasibility and generated critical customer feedback, enabling a startup serving the fashion industry to pivot before expensive full development. Rapid prototype-driven learning saved months of building the inappropriate product. (startup in stealth mode)
Analyzed the impact of Energy Performance Certificates on residential property prices. The report provided quantitative evidence for choosing the most promising business model among three potential ones. Data-driven decision saved entrepreneur from pursuing low-viability opportunities. (entrepreneur in stealth mode)
Built an interactive prototype using open data to identify high-opportunity geographies for CO2 recycling, enabling data-driven market prioritization. Replaced gut-feel expansion strategy with evidence-based targeting of regions with highest symbiosis potential. (startup in stealth mode)
Research Prototypes
Generated knowledge graphs from diverse geographic data sources to enable faceted exploration of image search results [POI Exploration | WWW ’10] (Yahoo! Labs / Yahoo! Image Search).
Built a prototype for facilitating exploration of Flickr photos through semantically structured tag clouds [TagExplorer | technical report] (Yahoo! Labs).
Built a prototype providing focused access to sub-article-level information from Wikipedia through a specialized search engine [Wikiii | DIR 2006] (University of Amsterdam)