# Case studies

All case studies by Phil Diegmann: data pipelines, LLM evaluation and agile delivery.

- [FeedForward: formative feedback for open-ended process modelling tasks](https://dgmn.eu/work/feedforward-llm-feedback/index.md): A SolidJS web application for individual feedback on open-ended BPMN tasks, without grading by AI. Deterministic checks run first, the AI suggests and phrases, and the instructor decides every finding that reaches a student. (Designer and developer, 2026, University of Bremen (ISBPM seminar))
- [LLM judges for BPMN process model annotation](https://dgmn.eu/work/bpmn-llm-judges/index.md): LLM judges label a process model dataset, but their labels only become training data after agreement with human experts is measured. (Pipeline design and validation, In progress, 2026)
- [Trace data pipeline for open-source coordination research](https://dgmn.eu/work/trace-data-pipeline/index.md): 8 open-source projects, 1,169,489 actions, 93 major releases, prepared as time-ordered event sequences. Published in MIS Quarterly (2024). (Data infrastructure, co-author, 2017 to 2023, Cologne Institute for Information Systems)
- [Pseudonymised linking of survey and trace data](https://dgmn.eu/work/telecom-pseudonymised-linking/index.md): A hashing and pseudonymisation scheme that links survey responses to the same team's Jira and GitLab trace data without re-identifying individuals. (Pseudonymisation and data linking design, In progress)
- [Android migration of a B2B task-management and location-tracking app](https://dgmn.eu/work/android-migration/index.md): Led the migration of a B2B task-management and location-tracking app from Windows Embedded CE to Android, rolled out to several hundred devices. The team grew from one developer to four; I onboarded, reviewed and paired with the new developers and introduced Scrum. (Migration lead; onboarding and mentoring of new developers; introduced Scrum, 2012 to 2017, blueworld GmbH)
- [Self-hosted Kubernetes platform across home and cloud](https://dgmn.eu/work/kubernetes-platform/index.md): A Talos Kubernetes cluster across home nodes and a cloud VPS, with single sign-on, DNS, tunnels and encrypted secrets, all managed as code. (Owner, platform design and operation, Ongoing)
- [Reproducible team metrics from GitLab for project courses](https://dgmn.eu/work/dip-metrics/index.md): Same snapshots in, byte-identical scores out, and every ambiguous rule decision recorded. The lecturer keeps the grade. (Developer, 2026)