Description

I build AI systems end to end: problem scoping, data, model training, evaluation, and production delivery. If you need an LLM pipeline, a machine

learning system, or a data product that works in practice, I can own the full technical chain from day one.

What I bring to your project:

- LLM fine-tuning, RAG pipelines, and agentic AI systems deployed over live data streams

- Foundation model training and benchmarking on datasets of 1.2M+ records

- Federated learning across distributed networks with privacy and governance controls

- Adaptive ML for real-time IoT and streaming data

- Causal inference and experimental design for rigorous measurement

I hold a PhD in Computer Science and have published research across LLM systems, adaptive IoT ML, and federated learning. I have delivered results to

clinical, engineering, and product teams and am experienced explaining technical decisions to non-technical stakeholders.

I work best when I own the problem and can move fast. Comfortable with ambiguity. Focused on measurable outcomes.

Available for project-based work, short retainers, and technical advisory in LLMs and GenAI, ML pipeline design, AI architecture, and clinical AI

applications.

Industry field of expertise

Languages

English
Fluent
Dutch
Basic
Persian
Native or bilingual

Workplace preferences

Can work on-site

Amsterdam (up to 50km)

Amsterdam UMC,
Postdoctoral Researcher
January 2024 - Today (2 years and 6 months)
Netherlands
Owned end-to-end development of ML and language model pipelines on 1.2M+ clinical records: data preparation, feature engineering, model training, benchmarking across architectures, and delivery to clinical teams, improving balanced accuracy from 72% to 85% and macro-F1 from 70% to 78%.
Designed novel representation learning methods for clinical tabular data: label-free correspondence scoring for latent quality assessment, interpretable latent biomarker discovery, and attribution-based rule generation bridging deep learning and clinical interpretability. Built causal generative models over lipidomic datasets to produce high-fidelity synthetic training data, enabling model development and validation under strict patient privacy constraints. Architected and deployed a federated learning framework across 3 hospitals (Netherlands, London, Cambridge) within a 30-hospital consortium: privacy-preserving distributed model training, GDPR compliant governance controls, and live monitoring across nodes. Supervised 3+ junior researchers: structured feedback, code review, experiment design, and evaluation methodology.
University of Groningen,
PhD Researcher
January 2020 - January 2024 (4 years)
Netherlands
Designed and deployed a distributed LLM agent system for structured natural language output generation with an iterative self-refinement loop over live data streams; raised detection accuracy from 55% to 80% through prompt engineering and agentic evaluation loops (ACM DEBS 2024). Built AQUA-CEP, an adaptive ML system for real-time IoT sensor data streams that monitors data quality continuously and triggers automatic model retraining on distribution shifts (ACM DEBS 2023, 16 citations). Designed App-CEP, a real-time recommendation system with user-level privacy controls at the pattern level, published in Elsevier Information Systems Designed and ran controlled experiments across all projects: defined hypotheses, built evaluation frameworks, measured outcomes statistically, and used results to drive deployment and roadmap decisions. Owned the full lifecycle of all research projects from problem formulation through production deployment and iteration.