Mohamed Rodani

Mohamed Rodani

Physics @ QMUL • Machine Learning, Quant Finance & Scientific Computing

Turning messy problems into clear models and readable explanations. Rigor-first approach to building strong fundamentals through consistent learning and hands-on practice.

Physics Machine Learning Quant Finance Scientific Computing Rigor-first Open-source / GitHub
About Me View Projects Let's Connect
View CV Download CV

Quick Facts

Education Physics @ QMUL
Interests
ML Quant Scientific Computing
Currently Deepening fundamentals + building portfolio
Open to Internships + collaborations
Featured Projects 0+ projects Competition Results 1st, 2nd & Top 25% ML & Quant Focus Python, PyTorch

About

I'm a Physics student at Queen Mary University of London (QMUL), working at the intersection of mathematics, physics, machine learning, and quantitative finance. I build end-to-end ML projects with rigorous evaluation and reproducible results.

  • Focus: End-to-end ML systems, financial sentiment analysis, and quant frameworks
  • Build: From algorithms (backprop from scratch) to production apps (full-stack ML)
  • Practice: Kaggle competitions, hackathons, and collaborative projects through QMUL ML Society
  • Seeking: ML/quant spring weeks and internships for 2026–2027

Highlights

QMML Market Making Hackathon - 2nd Overall & 1st Sortino

2nd place out of 93 teams in a 9-round live market-making competition (£305,727 final cash); best risk-adjusted returns (Sortino) in the field

QMUL ML Society Valentine's Hackathon - 2nd Place

2nd place finish (0.59235 AUC, 1st: 0.59275) predicting Valentine's date likelihood using a rank-based ensemble of LightGBM, XGBoost and CatBoost tuned with Optuna

QMUL ML Society Christmas Hackathon - 2nd Place

2nd place finish predicting customer return behaviour using strategic feature engineering (17 features beat 55+ one-hot encoded)

Kaggle Top 25% - Loan Default Prediction

Ranked #950/3,850 in S5E11 with 0.92450 AUC using LightGBM + Optuna optimization

BERT Financial Sentiment Analysis

Fine-tuned BERT model achieving 87% test accuracy with improved negative sentiment recall

Projects

Featured Projects

Selected deep dives — longer write-ups and reproducible repos

QMML Market Making Hackathon - 2nd Overall

  • Nine-round live trading competition (QMUL Machine Learning Society): model a hidden price from features, submit bid/ask quotes, then trade against the market maker.
  • 2nd of 93 teams with £305,727 final cash (~3× starting bankroll); 1st on Sortino for risk-adjusted returns.
  • Tiered sizing by model confidence, deliberately avoided the market-maker role, and used Kelly-style risk discipline.
2nd / 93 teams 1st Sortino 9 live rounds
Case Study GitHub
Performance Vectors chart — competitor bankrolls across hackathon rounds R2–R9
Python Machine Learning Scikit-learn Quantitative Finance SciPy

Base Gas Optimiser - Hackathon Winner

  • Predicts Base network gas prices with scikit-learn to help users save on transaction fees.
  • Full stack: React/TypeScript frontend, Python Flask backend, and live on-chain data.
  • Live gas indicators, 24-hour forecasts, and a savings calculator with MetaMask support.
Hackathon Winner Up to 40% gas savings Full-stack ML app
Case Study GitHub
Base Gas Optimiser - Gas Price Prediction Dashboard
Python Flask React TypeScript Scikit-learn Blockchain

BERT Financial Sentiment Analysis

  • Fine-tuned BERT (110M parameters) on financial news headlines: positive, negative, and neutral.
  • Balanced F1 across sentiment classes.
  • Raised negative-class recall from 28.6% to 85.7% using weighted loss and early stopping.
87% test accuracy 85.7% negative class recall 5,842 headlines
Case Study GitHub
BERT Financial Sentiment Analysis - Confusion Matrix and Performance Metrics
Python PyTorch Transformers NLP FinBERT

QMUL ML Society Valentine's Hackathon - 2nd Place

  • 2nd place: predict whether someone has a Valentine's date from demographic and social attributes.
  • Target encoding with KFold CV, feature interactions, and log-transformed income.
  • LightGBM, XGBoost, and CatBoost tuned with Optuna (50 trials each), combined in a rank-based ensemble.
2nd Place 0.59235 AUC 700K training rows
Case Study GitHub
Python LightGBM XGBoost CatBoost Optuna Feature Engineering

QMUL ML Society Christmas Hackathon - 2nd Place

  • 2nd place: predict holiday-season customer return behaviour.
  • 17 curated features (aggregation, temporal, core signals) vs baselines with 55+ one-hot encodings.
  • Simple LogisticRegression — strong features mattered more than model complexity.
2nd Place 8,000 transactions 17 curated features
Case Study GitHub
Python Scikit-learn Feature Engineering Classification

More Projects

Additional projects and ongoing work

RYM Investment Framework

  • Quantitative investment framework for UK Investment Challenge 2025.
  • Systematic portfolio construction with risk analysis and performance metrics.
  • End-to-end data pipeline for quantitative asset analysis.
Case Study GitHub
RYM Investment Framework
Python Quantitative Finance Data Pipeline

Kaggle Playground Series - Top 25%

  • Monthly Kaggle Playground Series entries; November 2025 (S5E11): Loan Default Prediction.
  • Top 25% (#950 / 3,850) with 0.92450 AUC.
  • LightGBM + Optuna, multi-seed averaging, 593,994 training samples.
Top 25% (Nov 2025) 0.92450 AUC 593K samples
Case Study GitHub
Kaggle Playground Series Overview
LightGBM Optuna Kaggle Classification

ML Mini Projects

  • Focused ML experiments: backprop from scratch (NumPy), regression, classification, and related fundamentals.
  • Each mini-project stresses the mechanics behind common algorithms.
Machine Learning Mini Projects
Python Machine Learning Data Science

Heat Pump Explained

  • Educational hackathon project: heat pump technology explained for non-technical audiences.
  • Physics, computational modelling, and interactive visualizations for core thermodynamics ideas.
Case Study GitHub
Heat Pump Explained
Python Physics Data Visualization

Experience

March–April 2026

QMML Market Making Hackathon - 2nd Overall

  • Competed in a 9-round live market-making challenge with 93 teams; each round used tabular features to model a hidden asset price, then submit quotes and directional size
  • Finished 2nd overall with £305,727 final cash (~3× starting £100k) and won 1st place on Sortino ratio for best risk-adjusted performance
  • Strategy: avoided being market maker (wide non-competitive quotes), tiered models by validation strength (linear models where R² was high, regularisation / baselines where signal was weak), and Kelly-motivated sizing with survival constraints
  • Artifacts: Case study · GitHub (notebooks, strategy scripts, Kelly / quote tools)
February 2026

QMUL ML Society Valentine's Hackathon - 2nd Place

  • Predicted whether a person has a Valentine's date from demographic and social attributes on a 700K row dataset, evaluated on ROC-AUC
  • Engineered features from Survey_Date (month, hour, day), created interaction terms, log-transformed income, and added missingness flags
  • Applied target encoding with KFold cross-validation on categorical variables to prevent data leakage
  • Tuned LightGBM, XGBoost, and CatBoost with Optuna (50 trials each) for thorough hyperparameter optimisation
  • Combined all three models with a rank-based ensemble, achieving 0.59235 AUC (1st place: 0.59275)
  • Artifacts: Case study · GitHub
December 2025

Coinbase Hackathon 2025 - Winner

  • Built full-stack ML application for Base network gas price prediction in a team, helping users optimize blockchain transaction fees through real-time price forecasting
  • Developed scikit-learn models integrated with React/TypeScript frontend and Python Flask backend for seamless user experience
  • Implemented real-time blockchain data integration with live gas indicators and 24-hour price predictions
  • Created savings calculator with MetaMask wallet support, enabling users to quantify potential cost savings
  • Delivered production-ready solution demonstrating end-to-end ML pipeline from data collection to deployed application
  • Artifacts: Case study · GitHub
December 2025

QMUL ML Society Christmas Hackathon - 2nd Place

  • Predicted holiday-season customer return behaviour using 17 curated features (aggregations, time, core signals); beat a 55+ one-hot baseline with a simple LogisticRegression pipeline
  • Artifacts: Case study · GitHub
November 2025

Diverse AI Hackathon 2025

  • Built AI transparency engine for smart heat pump systems in a 5-person team, addressing trust barriers in home energy automation through pattern detection and natural language explanations
  • Developed rule-based classifier identifying 5 operational scenarios (tariff optimisation, VPP curtailment, predictive heating, efficiency anomalies) with 3-tier recommendation system
  • Integrated Claude Sonnet 4 API with <50 word constraint, converting technical metrics (COP, flow temps, grid signals) into actionable insights
  • Grounded solution in EoH dataset (742 UK installations, 2+ years monitoring data) for realistic COP/SPF simulation
  • Delivered Streamlit dashboard with live Q&A chatbot, data transparency features, and benefit quantification (£ + kg CO2)
  • Artifacts: Case study · GitHub
2025 - Present

QMUL Machine Learning Society - Quant Division

Active Member

  • Active participant in the quantitative finance division — trading ideas, notebooks, and collaborative builds
  • Example build: QMML market-making hackathon (see Projects for other society-related work)

Education

2025 - 2029 (Expected)

Physics BSc — Queen Mary University of London

Bachelor of Science (Honours) in Physics

Undergraduate physics at QMUL, developing mathematical modelling, computational methods, problem-solving, and analytical thinking alongside independent ML and quant projects.

2022 - 2024

A Levels

Mathematics, Further Mathematics, Physics

Completed A Levels in Mathematics, Further Mathematics, and Physics, building a strong foundation in quantitative and analytical skills.

2022

GCSEs

4 Grade 8s, 3 Grade 7s, 2 Grade 6s

Completed GCSEs with strong results across multiple subjects.

Skills

Proficient Comfortable Learning

Languages & Tools

Python NumPy Pandas Matplotlib Seaborn Scikit-learn Git Jupyter SciPy PyTorch XGBoost LightGBM TensorFlow

Machine Learning

Supervised Learning Tree-Based Models Deep Learning NLP & Transformers Feature Engineering Time-Series PINNs

Mathematics & Theory

Linear Algebra Calculus Probability & Statistics Differential Equations Mathematical Modelling

Contact

Seeking ML/quant spring weeks and internships for 2026–2027. Open to ML hackathons and collaborative projects. If you're recruiting in machine learning, quantitative finance, or running ML-focused events, I'd love to connect.

GitHub LinkedIn

Contribution chart

My contribution activity

GitHub Contributions Chart for M-Rodani1