Machine Learning for Economics
Teaching Assistant · University of Bergamo. Introduction to the R programming language.
Ph.D. Candidate in Economics · University of Bergamo
Mohsen Rahimi
I study how political institutions and public decision‑making shape economic outcomes — using text‑as‑data, large language models, and interpretable machine learning as measurement tools.
01
About
I am a Ph.D. candidate in Economics at the University of Bergamo, advised by Sergio Galletta (ETH Zürich / Sapienza University of Rome). My work sits at the intersection of empirical political economy and machine learning: I build measurement tools from text — central‑bank transcripts, audit reports, policy documents — and use them to ask how institutions actually make decisions, and for whom.
Before the Ph.D. I worked as a data scientist in Milan, where I designed explainable‑AI systems for NLP models and deployed production ML pipelines on Kubernetes and Google Cloud. That experience shapes how I approach research: models should be transparent enough to be audited, and fair enough to be trusted with policy.
I hold an M.Sc. in Economics and Data Analysis from Bergamo and a B.Sc. in engineering from the University of Tehran.
02
Research
Working papers and publications. Click a title to read the abstract.
-
Using large language models to analyze Federal Open Market Committee transcripts, this paper moves beyond basic sentiment to measure features of deliberation — including speaker responsiveness and directional convergence — and asks whether the quality of a meeting's discussion predicts the accuracy of the Fed's subsequent economic forecasts.
-
Building on Ash and Galletta's machine‑learning approach to anti‑corruption policy, I analyze how using ML to target public audits in Brazil creates unintended regional disparities in exposure. I introduce a fairness‑constrained optimization step that maintains high corruption‑detection rates while correcting unequal geographic exposure.
-
Interpretability is essential for trustworthy NLP: it helps identify biases and errors and improve model performance. We propose an interpretability system that analyses and interprets the predictions of black‑box NLP models using adversarial examples, combining local and global interpretability methods for a more comprehensive understanding of model behaviour — and, in particular, of when and why models fail.
Rahimi, M., De Poli, G., Masella, A., & Bregonzio, M. (2023). Why did you fail? An interpretability system for NLP models. In D. Kamissoko et al. (Eds.), Proceedings of the 9th International Conference on Decision Support System Technology (ICDSST 2023): Decision Support System in an Uncertain World — the Contribution of Digital Twins (pp. 199–205). IMT Mines Albi, France.
Author lists on working papers reflect the current state of each project.
03
Experience
-
2022 – 2023
Data Scientist · Datrix AI Solution Group (3rdPlace), Milan
- Developed an adversarial machine‑learning framework for NLP, combining SHAP‑based local explanations with global model insights to reduce bias and interpretation variability.
- Built an explainable‑AI module (AIA Guard) for the Italian Government's CybersecH project.
- Led cloud deployment of high‑performance ML pipelines on Kubernetes and Google Cloud.
-
2021 – 2022
Scientific Research Assistant · University of Bergamo
- Exploratory analysis and cleaning of ISTAT insurance microdata; contributed to a customer‑retention optimization algorithm.
Awards & honors
2021
1st Place, Stat‑Hackathon
University of Bergamo. Fraud detection on UniCredit bank transactions, with Capgemini, A2A Smart City, IN.TWIG and OROBIX.
2021
Top‑10 Finalist, Future of Mobility Program
With Harvard Graduate School of Design. Ranked 9th of 100+ participants for an agent‑based simulation of disruption behavior.
04
Teaching
Basics of Programming Languages
Teaching Assistant · Faculty of Engineering, University of Tehran.
C++, Python & Mathematics
Teacher · Salam High School, Tehran.
05
Toolkit
- Programming
- Python, R, Stata, Java, C/C++
- Economics
- Econometrics, causal inference, policy modeling, behavioral analysis
- ML / AI
- PyTorch, TensorFlow/Keras, HuggingFace, SHAP, scikit‑learn, H2O, tidymodels
- Data & infrastructure
- PostgreSQL, Neo4j, Docker, Kubernetes, Google Cloud, Git, Linux, LaTeX
- Languages
- Persian (native), English (IELTS 7.0), Italian (CILS B2)
06
Get in touch
Happy to talk about political economy, text‑as‑data, or anything at the boundary of economics and machine learning.
GitHub
@mohsenrahimi13
LinkedIn
in/rahimimo
Office
Department of Economics
University of Bergamo, Italy
University of Bergamo, Italy