Curriculum Vitae

Doctoral research in biosignal data analysis and medical AI. Focus on machine learning approaches for intraoperative patient monitoring and hypotension prediction.

Specialisation in Macromolecules and Complex Systems. Thesis: Eignung von rekurrenten Deep Learning Methoden zur Analyse von Biosignalen, AG Cardiovascular Physics.

Thesis: Untersuchungen zum Dotieren von Poly(3-hexylthiophen-2,5-diyl) am Beispiel eines organischen und eines anorganischen Akzeptors, AG Supramolecular Systems.

Research data scientist on the TRANSFER project - an end-to-end ML system for intraoperative patient monitoring. Built the data infrastructure: ingesting OR time-series streams via Kafka, persisting to TimeScaleDB and a FHIR store, running preprocessing and inference through a Triton Model Server, and deploying all modules as Docker containers. Designed and trained neural network models for near-real-time intraoperative hypotension prediction across multiple severity classes, plus statistical clustering and rule-based classification for hemodynamic endotypes. Preliminary models trained on the ~6,000-case VitalDB public dataset, scaling to an incoming ~300,000-case Charité clinical cohort.

Developed Revenue Group Predictor - a Python REST webservice returning probability distributions over patient length-of-stay intervals. Refactored the text-transform and preprocessing pipeline, retrained the model on a ~10M-record dataset, and migrated model serving to ONNX.

De facto go-to person for AI/LLM tooling across the company: evaluating AI developer tooling for org-wide adoption - including Claude Code, OpenCode, the uv Python package manager, and open-source MCP servers connecting agents to Jira, Confluence, and a self-hosted TFS instance. Designed and delivered internal AI training sessions (60–80 attendees per session, ~100 of 120 colleagues reached) and mentors colleagues adopting these tools.

Anomaly detection on EV charging station data: meter-jump analysis, POWER METER FAILURE pattern investigation, correlation between log histories and station outages, and timeout clustering.

Developed the Status Cycle metric - a silent-fault indicator for chargers oscillating spontaneously between AVAILABLE and PREPARING states without issuing an error. Built a Grafana monitoring dashboard (CDR Gap, Status Cycle, WEAK SIGNAL, error alerts) across all stations with automated Zammad ticket dispatch and per-station time-series views.

Extended the GIS plugin with municipal boundaries, public on-street parking, and Census 2022 data. Delivered ATLAS, a next-generation GIS PoC for expansion potential analysis (Leipzig pilot). Led a Smart City Systems sensor test as project manager, integrating the device via the ParkingPilot API.

Sole developer of a PyQGIS/PyQt GIS plugin for EV charging site scouting, integrating OpenStreetMap and LEMNET; first version deployed in three weeks. Scouted ~200 potential locations in 6 months with an 80% permit approval rate. Co-managed permit applications, exceeding the 160-station target (170 secured, 25% of Berlin's total allocation). Monitored construction projects (~€12,000 budget each) and commissioned and maintained 22 charging stations.

Led operations for outdoor construction, landscaping, removals, and winter services. Managed ~15 client accounts (Vonovia, Covivio, and others), coordinated teams of up to 20, and oversaw procurement, project closure, and bookkeeping (~€1.8M annual revenue).

Tenant relations for 86 residential units; bookkeeping (~€750K annual revenue).

Lab assistant at AG Physik von Makromolekülen: experiment execution, chemistry lab work, and electron microscopy.

EU-wide consultant acquisition for the e-Health Hub; onboarded 120 consultants from all EU member states.

ECG acquisition and analysis app built in Flutter. Uses a Polar H10 chest strap over Bluetooth to record 5-minute ECG sessions, uploads to a Google Firebase backend, and runs a signal processing pipeline (noise reduction, beat detection, HRV & cardiac parameter calculation). Documented MDR regulatory classification and compliance path for a Class IIa medical device.

Agentic LLM framework that closes the loop between project manager, developer, and tester roles. Automates iterative software development tasks using multi-agent coordination.

SKU product lifecycle management tool. Manages the full lifecycle of product stock units from creation through end-of-life using structured LLM-assisted workflows. Agentic LLM component queries supplier and compatibility databases via tool use to autonomously improve component suggestions to the user.