Model evaluation
Accuracy, calibration, and error-pattern checks that decide where a model can run on its own and where it needs review. It's what gets a model trusted in production.
Quantitative Analyst · Economics PhD
Jesper
Armouti-Hansen
Quantitative analyst with an economics PhD and a decade of applied data work. I build statistical and machine-learning models on structured data, and assess when their output is reliable enough to act on.
Cologne, Germany · Python · machine learning · econometrics
Profile
What I do
Statistical & econometric modeling
Statistical and econometric models, built and estimated on structured data, with a clear read on how much of the variation they actually capture.
Machine learning on structured data
Machine learning on tabular data (gradient boosting, regularized models, and the like), always benchmarked against a simple baseline so the added complexity has to earn it.
Reproducible code
Version-controlled, tested code in Python, SQL, and R. The economics training is the edge on the harder part: telling a real effect from noise.
Selected work
Analysis, models, and checks
A few projects, each with the question it addresses, the method, and the result.
RAG Search Engine
A hybrid movie-search engine combining lexical and semantic retrieval, reranking, image search, and RAG answers.
- Problem
- Plain search fails when a query is vague, visual, or more description than title.
- Approach
- Combined BM25, embeddings, reciprocal rank fusion, CLIP, reranking, and caching, with explicit checks at each retrieval stage.
- Result
- A pipeline with each retrieval stage measured separately, so ranking quality is attributable to a stage rather than the whole.
Python · RAG · BM25 · embeddings · CLIP · evaluation
CitiBike Demand, Risk, and Net Flow
Station- and trip-level analysis of demand, net flow, and collision-adjusted risk across New York's bike-share network.
- Problem
- Where does the network strain — heavy use, poor rebalancing, or higher risk per trip?
- Approach
- Joined trip records with collision data and built station-level measures that separate raw volume from exposure-adjusted risk.
- Result
- Three separate measures — demand, imbalance, and exposure-adjusted risk — that a single ranking would have conflated.
Python · pandas · feature engineering · risk analysis · forecasting
Economic Theories and Machine Learning
Code comparing economic theories against machine-learning benchmarks on behavioral data.
- Problem
- When a theory predicts behavior, how much of the predictable variation does it actually capture?
- Approach
- Benchmarked theory-based specifications against machine-learning models and examined the remaining predictive gap.
- Result
- Theory specifications scored on out-of-sample prediction and compared against the machine-learning benchmark.
Python · machine learning · model evaluation
Experience
Where I've worked
Data Scientist, AXA Konzern AG
- Analyze model outputs and their confidence levels to decide when a prediction can be used as-is and when it needs human review.
- Set the review thresholds on that basis, cutting manual review by around 30%.
- Work with the model-owning teams to put those findings into production.
Postdoctoral Researcher in Economics, University of Bonn
- Applied econometric and statistical methods to behavioral and economic data.
- Built and validated empirical models, and quantified how much they actually explained.
- Published the results in peer-reviewed papers and presented them at conferences.
Background
Research & teaching
A decade in academic economics — published research, refereeing for journals, and university teaching.
4peer-reviewed papers
PhDeconomics, summa cum laude
5journals refereed for
60+theses supervised
Contact
Get in touch
Email is the best way to reach me — I'm glad to talk through a quantitative or modeling problem.