Sören Wacker

Designed a schema-first metadata governance system that generates Pydantic models from YAML specifications. Enforces ontology-backed validation across REST API, CLI, and web interfaces to ensure data quality at ingest boundaries.

• Role: Lead Architect
• Architecture: Factory pattern for entity generation, ontology URI resolution
• Interfaces: FastAPI REST, Typer CLI, HTMX web UI
• Governance: Schema validation prevents dirty data entering downstream ML pipelines
• Access: Login via SRAM account or register with EduID

Selected for the TDCC-LSH (Life Sciences & Health) FAIR Fellowship Programme (NWO-funded) to develop expertise in enterprise data governance, metadata standards, and AI-ready research data infrastructure.

• Role: FAIR LSH Fellow
• Mentor: Sharif Islam (Naturalis)
• Focus: Data governance, metadata standards, interoperability
• Duration: 12-month programme

Architected an observability pipeline that exports the SLURM database to DuckDB for historical analysis. Visualizes cluster utilization patterns to inform capacity planning and resource allocation decisions.

• Role: Lead Architect & Developer
• Scale: Multi-tenant HPC cluster with hundreds of users
• Architecture: SLURM database export, DuckDB analytics, Plotly.js dashboards
• Impact: Data-driven resource allocation and capacity planning

Architecting ResilienceHub, a federated research data platform spanning TUD, UvA, UU, and WUR. Designed multi-tenant identity management with Keycloak, container-orchestrated microservices, and MLflow integration for experiment tracking across distributed research teams.

• Role: Consultant for Systems Design, Architecture, and DevOps
• Scale: 4 universities, multiple industry partners, ~58M combined funding
• Architecture: Keycloak federated identity, S3 object storage, FairdomSEEK integration
• MLOps: MLflow experiment tracking, GitOps continuous deployment
• Data Governance: FAIR-compliant data management with FairdomSEEK

Delivered browser-based computer vision POC in 6 weeks, demonstrating PyTorch model deployment for single-cell segmentation and classification. Architecture designed for zero-infrastructure inference to support grant application.

• Role: Solution Architect & Developer
• Constraint: 6-week delivery timeline
• Architecture: Browser-based inference, no server-side compute required
• Impact: Supported successful $2M Bezos Earth Fund grant

Led data architecture for the PRIME collaboration between UCalgary and Broad Institute. Managed multi-tenant database partitioning for genomics, proteomics, and metabolomics data linked to clinical records across multiple research groups.

• Role: Team Lead Data Science
• Scale: 35,000+ samples, 100TB+ data, multi-institutional collaboration
• Architecture: Multi-tenant data partitioning, HPC compute orchestration
• Compute: ~$420K DRAC grants for cluster allocation (applied by me)
• Publications: Nature Communications, Analytical Chemistry

Architected a distributed QC platform to replace inconsistent manual workflows. Designed job queue durability via Redis/Celery with container isolation to ensure zero data pollution across concurrent analyses in multi-tenant environments.

• Role: Lead Architect & Developer
• Problem: Inconsistent manual QC across research team
• Architecture: Django orchestration, PostgreSQL persistence, Celery/Redis job queues
• Isolation: Docker containerization for reproducible, isolated processing
• Impact: 98% reduction in QC processing time
• Publication: 'LAMPrEY: a Python-based automated quality control tool for large-scale proteomics datasets', bioRxiv (2026)

Built a scalable metabolomics processing engine after existing platforms failed at 3000+ file batches. Evolved from Jupyter library to standalone web application with interactive visualization and batch processing capabilities.

• Role: Lead Architect & Developer
• Scale: 3000+ file batch processing where competitors failed
• Evolution: Python library to Plotly Dash web application
• Architecture: Pandas data engine, parallel processing, interactive dashboards
• Publication: 'MS-MINT: An Open-Source Data Analysis Software for Large-Scale Metabolomics Studies', Analytical Chemistry (2026)

Co-founded ML startup with Li Ka Shing Institute to deploy predictive models for drug safety assessment. Built production ML pipelines for cardiac toxicity (hERG blockade) prediction to accelerate drug candidate screening.

• Role: Co-founder & ML Engineer
• Domain: Cardiac toxicity (hERG blockade) prediction
• Models: XGBoost ensemble methods for molecular property prediction
• Publication: 'Performance of ML algorithms for hERG1 channel blockade prediction' (2017)