AI Engineer & Applied Data Scientist
Utrecht, Netherlands
RAG & context engineering · Agentic data science · Operations research
My background combines Data Science research and AI engineering. My PhD research at Utrecht University required connecting scientific literature, mathematical methods, modelling assumptions, experimental findings and code. I developed research AI workflows to support that work, alongside probabilistic modelling of dynamic interactions.
My work focuses on:
- AI engineering and research knowledge: RAG, hybrid retrieval, reranking and context engineering, with knowledge-graph exploration and evaluation of source coverage, citations and unsupported answers.
- Agentic data science and experimentation: AI-assisted task decomposition, analytical code development and execution, simulation and model comparison, with computational checks and researcher review.
- Supporting software and reproducible workflows: Python applications, APIs and data pipelines, with FastAPI, AWS, Docker, Databricks and MLflow across research and independent projects.
Before my PhD, I worked on ecommerce forecasting, commercial planning and supply-chain optimisation at Digikala.
- JobLab Agent API: FastAPI job-search and CV-matching backend with a custom Bedrock tool-calling loop, MLflow tracing and Lambda deployment.
- JobPilot: Job-application tracking with Kanban boards, Nova task assistance, Gmail integration and a separate search backend.
- JobLab Lakehouse: PySpark transformations, data-quality checks and reproducible analytical tables across Databricks, Snowflake and Tableau.
- JobLab Data Pipeline: Job collection, skill extraction, standardisation and deduplication, with scheduled AWS Fargate infrastructure.
- JobLab Analytics: Interactive job-market charts, search, AI chat and CV matching in Next.js.
- MLflow Tracking Server: Containerised experiment and trace tracking, with PostgreSQL metadata and optional S3 artifacts.
- Tax Authority RAG: Assessment implementation using FastAPI, access-filtered retrieval, bounded LangGraph workflows and optional Bedrock integration. The separate architecture and evaluation repository contains the earlier design and mock-based test work.
- 2sHMMREM: R simulation and analysis workflows combining Hidden Markov Models and relational event models, with uncertainty diagnostics and repeatable reporting.
- RepairShop MILP: Repair-shop scheduling across GAMS/CPLEX, MATLAB NSGA-II and Python OR-Tools, balancing vehicle flow time and station costs.
- ANN-SMB simulation optimisation: MATLAB research comparing neural-network and Kriging metamodels for simulation optimisation and inventory control.
- Retrieval and context: connecting scientific sources, choosing useful context and examining where evidence is lost.
- Agentic applications: tool calling, bounded workflows, evaluation and reviewed outputs.
- Applied data science: probabilistic models, dynamic networks, forecasting, simulation and optimisation.
LLMs, RAG and agents
Development, infrastructure and cloud
Data science, optimisation and analytics
- Research tools that connect literature, methods, experiments and code.
- AI-assisted applications for search, analysis and workflow automation.
- Analytical models and software for forecasting, scheduling and supply-chain decisions.
In research, I need to understand the model and check the evidence behind the result. In commercial analytics, I need to connect that result to a planning or operational decision. My software and AI work supports both.
AWS Cloud Quest: Cloud Practitioner, Training Badge · Generative AI Practitioner, Training Badge