Machine Learning Engineer focused on reinforcement learning, deep learning systems, and performance optimization.
I enjoy building ML systems from the ground up — designing training pipelines, experimenting with architectures, and optimizing model performance at the systems level.
Building an agent-based ML system that ingests emails from multiple accounts and automatically categorizes and summarizes them.
The system organizes messages and generates summaries so important information can be understood without manually scanning entire inboxes.
Developing a personal tooling system to analyze GPU memory fragmentation during inference workloads.
The project focuses on improving performance and stability when running large language models.
Built a custom RL training environment for Factorio.
Implemented:
- custom mod for environment interaction
- action masking to prevent invalid actions
- adaptive epsilon exploration
- reward shaping targeting iron mining and smelting
The agent successfully learned the iron smelting production task.
Stack: PyTorch, reinforcement learning, environment design
Implemented a GPT-2 style transformer model trained on an author-specific dataset.
Built a custom preprocessing pipeline including:
- statistical TrueCase normalization
- custom tokenization pipeline
Stack: PyTorch, CUDA, transformer architectures
Developed a clustering pipeline to analyze customer behavior in multi-year retail datasets.
Stack: PySpark, Databricks, Polars
- Reinforcement Learning
- Agentic Systems
- Large Language Models
- Training Pipeline Optimization
- GPU Memory Management
- Simulation-based learning environments
Languages
Python • Rust • C# • Java • Kotlin • SQL • Bash
ML / Data
PyTorch • TensorFlow • JAX • NumPy • Polars • PySpark • Scikit-Learn
Tools
Git • Docker • AWS • Databricks
- RL agents interacting with complex simulations
- Systems-level ML performance optimization
- Agentic workflows for real-world automation