Machine Learning Engineer
I design and ship applied ML systems for multimodal search, NLP, and computer vision.
Experience across Adobe, BMW, and Aptiv, spanning data pipelines, model development, and production deployment on AWS.
Production-ready ML systems
Building retrieval systems, vision models, and reliable data pipelines with measurable impact.
Past Experience
BMW
Built multimodal retrieval models integrating text + structured data and developed scalable data pipelines on AWS Glue.
Barmer
Prototyped NLP models for extracting data from insurance PDFs with privacy constraints.
Adobe
Built image classification models for Lightroom and optimized deployment for mobile.
Aptiv
Implemented semantic segmentation for radar perception and evaluated neural architectures.
University of Wuppertal
Developed forecasting models for financial time series and built interactive learning tools.
University of Wuppertal
Analyzed hospital ER operations and simulated complex healthcare systems in Python.
Technical Expertise
Tools I use to move from data to deployed models.
PyTorch
Built end-to-end pipelines from research to production—custom architectures, self-supervised learning, and mobile deployment. Comfortable with the full stack: debugging training dynamics, optimizing compute graphs, and shipping models that actually run on phones.
Transformers & BERT
Pre-trained my own models on proprietary datasets and fine-tuned for domain-specific tasks. Know the difference between catastrophic forgetting and actually achieving transfer learning. Built retrieval systems that handle the scaling challenges nobody talks about (context length, vector DB bottlenecks, tokenizer mismatches).
Polars & Spark
Single-machine speed with Polars when it matters, Spark for problems that actually need distribution. Spent enough time optimizing ETL to recognize when you're fighting partitioning schemes instead of solving the real bottleneck. Handle terabyte-scale datasets comfortably.
MLflow & Experiment Tracking
Managed collaborative ML projects where you can't just train in notebooks. Know how to structure experiments so insights are reproducible, not scattered across Slack messages and lost runs.
SQL & DuckDB
Advanced query work—window functions, CTEs, recursive queries for real problems. DuckDB for when you need OLAP speed without the infrastructure tax. Comfortable writing SQL that's both correct and efficient at scale.
AWS Cloud
Production ML infrastructure: data lakes with S3, serverless ETL through Glue, compute on EC2. Infrastructure-as-code with Terraform so deployments are repeatable, not heroic.
Docker & MLOps
Containerized pipelines that actually work when someone else runs them.
Computer Vision & Mobile ML
Image classification, semantic segmentation, and the hard part—optimization. Reduced latency and memory footprint to make models viable for on-device inference.