AI / ML Engineer · NLP & LLM Specialist — based in Switzerland
I build machine learning and NLP systems that work in production — not just in notebooks. My focus is turning unstructured text and messy data into reliable, structured, measurable output, and making sure it runs dependably once it ships.
- 🧠 NLP & LLMs — named entity recognition, information extraction, text classification, RAG, and agentic workflows
- ⚙️ Production ML — training, fine-tuning, evaluation, and deployment of models that hold up under real conditions
- 🔬 Applied Data Science — classification, forecasting, and turning data into decisions
- 🚀 MLOps — pipelines, monitoring, and reproducible, maintainable systems
At John Snow Labs, I built production NLP systems for clinical, legal, and financial documents — NER and information-extraction pipelines with Spark NLP reaching 90%+ accuracy in production, with processing time cut by ~40%. I also contributed 67 merged pull requests to John Snow Labs' open-source repositories, including 10 to Spark NLP (4K+ ⭐), 44 to the Spark NLP Workshop, and 13 to johnsnowlabs, the gateway library to the John Snow Labs ecosystem. Most of my work over three years went into the company's private production repositories, which appear only as private contributions on my graph. Before that, I mentored a cohort of junior data scientists, guiding their ML projects from prototype to deployment.
Currently deepening my expertise in modern GenAI engineering — LLMs, RAG, agents, and evaluation — and starting a CAS in Advanced Machine Learning & MLOps at ZHAW.
Python · PyTorch · Hugging Face · Spark NLP · scikit-learn
LangChain · LangGraph · RAG · FastAPI · pgvector
PySpark · Pandas · Docker · CI/CD · AWS · GCP · Azure
| Project | What it is | Highlights |
|---|---|---|
| MCP Agent Tools | Model Context Protocol tool servers used by a LangGraph agent | stdio and streamable-HTTP servers · multi-turn memory · tests, CI and Docker |
| LangGraph Agents | Four agent design patterns, one project each | ReAct · Reflection · Reflexion · Agentic RAG with document and hallucination graders |
| Disease NER with Spark NLP | Disease mention extraction from biomedical text | BiLSTM-CNN-Char on the NCBI Disease corpus · token-level F1 0.84–0.85 · Streamlit demo |
| Slipway | Deployment blueprint: FastAPI on AWS with Terraform | 42 s to a live URL · 22 s teardown · OIDC-based CD from merge to live in 1 min 22 s |
| Retail Data Pipeline | On-premises ETL and analytics for 540K sales records | Validation with reject reasons · transactional, idempotent loads · Apriori and RFM · CI |
Three end-to-end platforms across healthcare AI:
- Clinical Intelligence Platform: RAG, structured extraction, agentic workflows and evaluation (LLM)
- ML Lifecycle Platform: end-to-end MLOps with training, serving, monitoring and retraining
- Healthcare Analytics Platform: data pipeline, predictive modelling and explainability



