Experience.

Kirk Larsen, a Danish law firm, runs its in-house AI applications on an on-prem Kubernetes cluster next to its own GPU servers. I am designing and building the Azure platform those applications are moving onto, one at a time, while the LLM fleet stays on-prem.

• 01Built the Azure platform from scratch as code: AKS provisioned with OpenTofu from an Azure DevOps pipeline, with state in Blob Storage behind Entra authentication and no storage keys anywhere. Every change goes through a pull request, and the apply runs exactly the saved plan.
• 02Set up the in-cluster platform: External Secrets backed by Key Vault through workload identity, cert-manager with Cloudflare DNS01, Traefik, Prometheus monitoring, and CloudNativePG with one database cluster per application and backups to geo-redundant Blob Storage.
• 03Connected the cloud cluster to the on-prem GPU models through a Cloudflare Tunnel with Access service tokens, so applications move to Azure without a configuration change.
• 04Rehearsed the first application migration against production data: restored its PostgreSQL cluster into Azure from its own backups, copied around 14,000 documents, ran the application in parallel, and wrote the cutover and rollback checklist.
• 05Keep a written decision log for the platform (OpenTofu over Terraform, pipeline deploys instead of ArgoCD, Key Vault over Sealed Secrets) with the reasoning and the cost of each choice, so the client's team can maintain and extend it.
• 06Led the root-cause analysis of an outage of the on-prem LLM fleet. Showed from metrics and filesystem evidence that two independent failures coincided: monitoring had gone blind hours earlier, while a rollout of large images filled the system disk. Restored disk headroom without downtime, recovered the CI builders, and recommended earlier alerting and clear alert ownership.

Modernised and operated the data and modelling platform behind real-time pricing for a global bunker fuel trading business, embedded in the client's cross-functional product team of developers, architects, QA and business stakeholders.

• 01Delivered a real-time pricing engine with statistical and machine learning models in the live serving path. Evaluated time-series approaches in Python and selected a Kalman filter on out-of-sample accuracy and fit to streaming data.
• 02Built the Python services, GraphQL and REST APIs and Dagster pipelines behind it: ingestion, validation, preparation and serving of real-time market data on PostgreSQL, with Dremio as a lakehouse query engine over Parquet-backed storage.
• 03Designed event-driven integration over Azure Service Bus between data pipelines, models and the operational trading systems, with explicit contracts, validation and traceability.
• 04Built and operated the production stack on Kubernetes with automated CI/CD and Terraform-managed infrastructure, and moved smaller services to Azure App Services where that made hosting simpler.
• 05Added data-quality checks, structured logging, monitoring and alerting (Dash0, Grafana, Sentry, Preset) that cut time-to-detection on pipeline failures from hours to minutes, and took part in on-call incident response.
• 06Introduced strict static type checking across the Python codebase with basedpyright, enforced in GitHub Actions so type errors failed the build instead of reaching production.
• 07Ran code review, set architectural standards, mentored colleagues and wrote the handover documentation, so the client's own team now operates and extends the platform.

Co-owned a medtech startup building a competency-assessment platform for surgical training. After each procedure the supervisor received a structured assessment of the trainee, and the aggregated results showed where each trainee needed more supervision. Deployed across multiple clinical departments under health-data compliance requirements.

• 01Built the backend in PHP/Laravel with MySQL, and the REST APIs behind the whole platform.
• 02Built the cross-platform mobile app in C#/.NET MAUI, used daily in clinical practice.
• 03Built the React front end, including the assessment workflow and real-time dashboards showing trainee progress and where supervision was most needed.
• 04Set technical direction, development standards, code review practice and documentation in a small team.

Long-running part-time engagement building and maintaining clinical data-management systems across more than 20 research projects, alongside my university role.

• 01Built custom REST APIs and workflow automation on top of REDCap, integrating data across research processes and external systems.
• 02Owned data structure design, validation, integration, documentation and compliance across the project portfolio.

Ten years in the prostate cancer research group at the Department of Molecular Medicine, developing non-invasive diagnostic and prognostic models from large-scale molecular and clinical data. Alongside leading my own research programme, I owned the group's software and data platform for its full lifetime.

• 01Developed and validated the uCaP, pCaP and MiCaP models: non-invasive diagnostic and prognostic signatures from cell-free material in urine and blood, taken from discovery through independent validation to prospective multicentre clinical trials.
• 02Modelled high-dimensional data with regularised regression (LASSO and elastic net) for feature selection, logistic and Cox proportional hazards models, and cross-validation with held-out and external validation cohorts, evaluated on ROC/AUC, sensitivity and specificity.
• 03Led the DPCC-1 study protocol, a multicentre prospective validation of the urine-based three-microRNA model uCaP across Danish hospitals, and co-invented a patent on microRNA-based early detection of prostate cancer in urine.
• 04Ran the PRIMA trial analysis comparing the STHLM3 test against PSA in general practice (European Urology Oncology), and designed and analysed a cluster randomised trial on how genetic risk assessment affects PSA testing in primary care (PLoS Medicine).
• 05Built and ran NGS pipelines on Illumina data (RNA-seq, CaptureSeq, QuantSeq, whole genome and cell-free DNA sequencing) on the GenomeDK Linux HPC cluster, plus SNP genotyping, imputation and eQTL analysis.
• 06Built the group's data processing, analysis and modelling tooling, mainly in C#/.NET alongside Python, R and SQL. It became the group's standard tooling and stayed in use after I handed it over.
• 07Automated the transfer of patient data from multiple REDCap databases into the prostate cancer biobank's SQL Server database in C# and PHP, replacing a manual process, and managed the biobank's data governance.
• 08Supervised PhD and MSc students, taught advanced courses, and coordinated EU-funded programmes, international collaborations and clinical trial compliance.
• 09Published 32 peer-reviewed papers, 9 as first author, with 1,119 citations and an h-index of 19.

PhD in the Laboratory of Genome Research on the role of DNA topoisomerases in transcription of chromatin-regulated genes. The work combined laboratory experiments with genome-wide data analysis, and it is where I started writing software: the questions were bigger than the available tools, so I built my own.

• 01Wrote a program that correlated our expression profiles against thousands of public microarray datasets, placing our results in the context of the wider literature instead of analysing them in isolation.
• 02Built a visualisation tool in C# that rendered flow cytometry DNA-content histograms as stacked ridgeline plots across a time course, so cell cycle progression could be read directly from the distributions. No off-the-shelf tool could produce it, so I wrote the rendering and layout myself.
• 03Applied statistical testing throughout: Pearson correlation with permutation-based significance, t-tests, Wilcoxon rank-sum tests and functional enrichment with multiple-testing correction, combining genome-wide screens with targeted ChIP-chip and qPCR.
• 04Published the work in PLoS Genetics and presented it at international conferences.

Completed military conscription and deployed with the Danish Army to Iraq, tasked with setting up and operating the communication network at the command support center.

• 01Ensured secure and reliable communications in a high-stakes conflict zone environment.
• 02Created a website in HTML so the squad could share photos with families back home, in 2005, before modern social platforms existed.