Production ML Engineer in Cape Town

Adrian Zevenster

I build and run machine learning systems for document intelligence, model evaluation, generative media, and operational decision-making.

Production ML, AI, and cloud engineering

My work covers Java and Python services, Vertex AI, Gemini, Document AI, BigQuery, Spanner, Kafka, ClickHouse, Terraform, evaluation, and production observability.

Selected work

  • RedStrings secure generative-media platform with isolated Google Cloud projects, IAP-protected manager and artist workflows, scheduled Drive intake, and exact-media provenance.
  • Multi-file KYC document verification with dedicated utility-bill, bank-statement, and financial-report intelligence paths.
  • A 19-case Gemini regression corpus with fixture, replay, live-model, and prompt-drift gates.
  • A floriculture data and forecasting platform with validated workbook ingestion, BigQuery and Dataform modelling, and nine-level reconciled forecasts on Vertex AI.
  • A full-stack data-analysis agent with 33 schema-validated tools, deterministic and LLM planning, hybrid retrieval, supervised ML, explainability, and feedback-driven evaluation.
  • Forecasting, churn, recommendation, migration, and real-time decision systems.

Read the engineering impact or browse selected projects.