CV#

Download as PDF

Data scientist and data science lead with 11+ years across industry and research. I build and evaluate LLM and machine-learning systems end to end, from benchmark design, rubric methodology and LLM-as-judge to the engineering needed to run them in production. A background in Bayesian statistics, causal inference and computational cognitive neuroscience informs an approach to LLM evaluation as a measurement problem: building evaluations that are reliable, interpretable and statistically sound. I have a track record of uncovering business-critical issues, leading cross-organisational technical work, and developing data scientists.

Experience#

Fab AI#

July 2025 – present

Head of Data Science#

August 2026 – present

  • Lead data scientist for the Quality Assurance Facility, Fab AI’s programme evaluating frontier and small language models for education in low- and middle-income countries.

  • Designed, with education experts, a benchmark suite testing whether LLMs can generate reading material matched to a learner’s phonics level, in English and Kiswahili; tech lead for two of them, directing a five-person team across two organisations.

Senior Data Scientist#

July 2025 – August 2026

  • Led the design, implementation and deployment (GCP) of an LLM-as-judge system scoring lesson plans, storybooks and textbooks against pedagogical rubrics. The rubrics and workflow are now used by external organisations to assess education technology.

  • Led rubric development with education specialists: binary criteria with fail conditions reached 87% human–LLM agreement against a 91.8% human–human ceiling, across GPT-5, Claude and Gemini.

  • Line-managed two data scientists: mentored one to a first-author publication, and designed a GCP benchmarking pipeline, built by the other, that automated previously manual model onboarding.

  • Introduced engineering rigour across the data science function: code review, project standards, Jira, and a move back to GCP.

eyeo#

March 2021 – March 2025

Staff Data Scientist#

March 2024 – March 2025

  • Uncovered a critical double-digit percentage user loss hidden by telemetry issues, directly impacting company strategy. Developed Bayesian hierarchical models to interpolate missing data, enabling accurate forecasting despite severe data quality issues.

  • Became the go-to data scientist for user retention and revenue analysis, presenting directly to C-level executives. Built causal models using DAGs to quantify the impact of retention initiatives.

  • Established company-wide data quality standards and analytical blueprints for causal inference. Aligned user metrics across three subsidiaries, creating the first unified view across programmatic advertising, browser, and B2B revenue streams.

Senior Data Scientist#

June 2022 – March 2024

  • Designed an ML-powered monitoring system that increased problematic ad detection by 580% while reducing manual review by 90%, improving the experience for 250M users.

  • Built data pipelines using Airflow, Docker, and GCP. Mentored 4 team members in modern data practices and led training sessions on Airflow, functional data engineering, and Bayesian modelling.

Data Scientist#

March 2021 – May 2022

  • Designed and productionised an ML model that increased problematic ad identification by 60%.

  • Built a Bayesian MRP model for estimating internet traffic.

  • Conducted strategic research and analysis for key business decisions.

University of Oldenburg — Researcher#

November 2016 – February 2021

  • Conducted research and published on the neural basis of speech processing using machine learning and human neural data, with a focus on interpretability and sparse coding.

  • Built research software (Python, Docker) for machine learning in neuroimaging that is still in use.

  • Supervised 5 Master’s research projects on machine learning applications in neuroscience.

Columbia University, New York — Visiting Researcher#

June 2019 – January 2020

  • Developed an interpretable deep learning model linking audio data to brain activity, based on self-attention in recurrent neural networks.

Self-employed / University of Oldenburg — Research Data Scientist#

November 2013 – November 2015

  • Statistical analyses in Python and R for Medical School Hanover and the University of Göttingen; modelling of non-linearities in human invasive neural recordings at Oldenburg.

Education#

PhD Candidate in Computational Cognitive Neuroscience (ABD)#

University of Oldenburg, 2016 – 2021

Completed all research requirements, including 2 peer-reviewed first-author publications; transitioned to industry.

MSc in Neurocognitive Psychology#

University of Oldenburg, 2013 – 2016

Skills#

General — Data science, research, software engineering, data engineering, stakeholder management

Programming — Python, SQL

Statistics — Bayesian and frequentist statistics, causal modelling, DAGs

Machine learning — Deep learning, (non-)linear regression and classification, unsupervised learning

LLMs — LLM-as-judge, evaluation and benchmark design, rubric development, structured output, multilingual evaluation

Software — NumPy, Pandas, scikit-learn, PyTorch, Airflow, Docker, Google Cloud Platform, PyMC

Selected publications#

See Publications for the full list.

Selected work#

See Projects for detail.

Extras#

  • Mentoring — I mentored junior data scientists at the University of Sussex.

  • Academic exchange grant (2019) — Awarded for developing a deep learning model for automatic speech recognition incorporating neuronal data.

  • Communication — I have presented my work at five conferences.