I recently completed a DPhil in Mathematical Physics at the University of Oxford. My background combines rigorous mathematical research with practical work in Python, statistical modelling, machine learning, optimisation and AI evaluation.
I enjoy turning open-ended technical problems into working models and tools. My projects span constrained optimisation, ML pipelines, time-series backtesting, simulation and interactive applications, and I also work professionally evaluating LLM outputs on advanced mathematics, physics and reasoning tasks.
I am interested in roles across applied AI, data science, quantitative modelling, research engineering and analytics, particularly where strong mathematical reasoning, programming and technical judgement are valuable.
Deployed Python application for modelling and optimising teams in the official F1 Fantasy game.
The project combines live and historical data, expected-value modelling and mixed-integer optimisation to select teams under budget, roster and transfer constraints. It includes data validation, race-weekend logic, configurable modelling assumptions and an interactive Streamlit interface.
Relevant skills: Python, pandas, PuLP, Streamlit, external APIs, constrained optimisation, data pipelines, expected-value modelling, testing and debugging
Live app: https://f1optimise.streamlit.app/
Repo: f1_fantasy_optimizer
End-to-end machine-learning pipeline for credit-default classification, covering data preparation, model comparison, cross-validation and decision-threshold tuning.
Models include logistic regression, random forest and XGBoost, with PR-AUC used alongside ROC-AUC to evaluate performance under class imbalance. Threshold analysis increased minority-class recall from approximately 35% to 59%, while making the corresponding false-positive trade-off explicit.
Relevant skills: Python, pandas, scikit-learn, XGBoost, cross-validation, model evaluation, imbalanced classification, threshold optimisation
Repo: ml_pipeline
Reproducible research and backtesting framework for investigating conditional market patterns using hourly financial data.
The project separates exploratory analysis from strategy evaluation and includes risk-based position sizing, layered exits, transaction-cost assumptions and performance and drawdown diagnostics.
Relevant skills: Python, pandas, time-series analysis, backtesting, performance evaluation, risk diagnostics, reproducible research
Repo: monday_range
Browser-based simulation of radar sweep detection, target acquisition and dynamic pursuit assignment.
Built as an interactive prototype for exploring decision logic in multi-target tracking and interception scenarios.
Relevant skills: JavaScript, HTML Canvas, simulation, real-time visualisation, decision logic, target assignment
Repo: drones
Programming & tools: Python, Git, Jupyter, Streamlit, SQL; working knowledge of JavaScript/HTML, MATLAB and C++
Python: pandas, NumPy, SciPy, scikit-learn, XGBoost, PuLP, Matplotlib
AI: LLM evaluation and failure analysis, prompt/task/rubric design, AI-assisted coding and research workflows
Modelling & analysis: probability and statistics, regression and classification, cross-validation, model evaluation, feature engineering, imbalanced classification, time-series analysis, simulation and constrained optimisation
Mathematics: linear algebra, probability and statistics, numerical methods, optimisation and stochastic processes
My doctoral research focused on mathematical structures in quantum field theory and categorical symmetries. I have authored or co-authored eight research papers, including work published in Physical Review Letters.
- LinkedIn: https://www.linkedin.com/in/daniel-pajer/
- Email: dancer-bout.6v@icloud.com