I am a postgraduate student currently reading for an MRes in Machine Learning and Big Data at Imperial College London, with a background in Theoretical Physics from UCL.

I characterise myself as a "high-slope" engineer: bridging the gap between complex mathematical modelling and practical, shipping software. I combine the analytical rigour of physics with the "get-it-done" mentality of a full-stack developer.

• 🔭 I’m currently working on: Advanced Computer Vision models and Reinforcement Learning agents.
• 💼 I have experience in: Quantitative Analysis, Full-Stack Web Development, and Commercial Risk Management.
• 🗣️ Languages: English (Native), German (Native/Fluent).
• ⚡ Fun fact: I used to play poker semi-professionally and worked as a dealer, which taught me everything I know about risk and probability.

Languages

Machine Learning & Data

Development & DevOps

• Tech: Python, PyTorch/TensorFlow, WGAN-GP, Geant4.
• Description: Engineering a Wasserstein GAN (WGAN-GP) to emulate 105 MeV pion-to-muon decay distributions for the COMET experiment. The model replaces computationally expensive Geant4 tracking in high magnetic fields by learning complex phase space manifolds.
• Impact: Enables high-statistics background modelling for BSM physics with a projected millionfold computational speedup over traditional methods.

• Tech: Python, FastAPI, Google Gemini (LLM), REST API.
• Description: Built a production-ready API that matches tenants to rental properties by inferring budget preferences from natural language conversations using Google's Gemini LLM.
• Impact: Processes 350+ properties and returns top 3 matches with price proximity within seconds. Deployed with full API documentation and testing suite.

• Tech: Python, TensorFlow, Keras, U-Net (CNN).
• Description: Engineered a Semantic Segmentation model to identify wildfires from satellite imagery. Handled a massive dataset of 51,000 images and solved severe class imbalance issues.
• Impact: Achieved 93.4% accuracy (Dice coefficient: 0.85).

• Tech: Python, Q-Learning, OOP.
• Description: Built a Reinforcement Learning agent trained on 250,000 real-money hand histories to master poker strategy.
• Impact: The model achieved a simulated win rate of +8 bb/100.

• Tech: Python, Time-Series Forecasting, Regression.
• Description: Developed an ML model to predict case numbers and death tolls across multiple time horizons (1, 2, 3, and 4 weeks ahead).

• Tech: Python, LSTMs (Long Short-Term Memory).
• Description: Used Deep Learning to predict the motion of a chaotic double pendulum system, demonstrating the application of LSTMs to complex physical systems.

Software Engineering Placement @ Sky

• Refreshed a legacy React component for an internal video tagging tool.
• Impact: Reduced content tagging time by 17% and resolved critical UI bugs in an Agile environment.

Poker Dealer @ Grosvenor Casinos

• Managed game integrity and financial risk in a high-volume cardroom with £1.2m yearly turnover.

• MRes Machine Learning & Big Data | Imperial College London (Current)
• BSc Theoretical Physics | University College London (UCL)
• Software Engineering Bootcamp | HyperionDev (Ranked 97th Percentile)

• Email: [email protected]
• LinkedIn: linkedin.com/in/jonathan-cassens-48967b225