Keyhan Kamyar

Building AI systems that learn, adapt, and perform under uncertainty.

Specializing in Deep Reinforcement Learning, ML Infrastructure, and High-Performance Computing. Eight years deep in the intersection of theory and production code.

🧠 What I Build

Reinforcement Learning Systems — Training agents that make sequential decisions in complex environments. From custom Gymnasium environments to production-grade RL frameworks optimized for financial markets.

ML Configuration & Experimentation — Type-safe frameworks for hyperparameter optimization, neural architecture search, and experiment tracking. Tools that eliminate boilerplate and enforce best practices from research to production.

High-Performance ML Pipelines — Python that runs fast. Custom Numba kernels, efficient data pipelines processing 100M+ row datasets, and optimization strategies that matter when milliseconds count.

Production ML Tools — Open-source libraries built from real research needs: configuration frameworks, financial data scrapers, proxy rotation systems, and infrastructure that bridges the gap between papers and deployment.

🎯 Current Focus

Developing novel RL architectures for algorithmic trading. Reformulating traditional actor-critic methods into more efficient multi-armed bandit frameworks. Building the infrastructure to train at scale.

Shipping open-source tools that other researchers and engineers actually use:

Type-safe search space configuration framework

High-performance financial tick data pipeline

Resilient proxy rotation for web scraping

🛠️ Stack

ML & RL: PyTorch • TensorFlow • Lightning • Gymnasium • Stable-Baselines3 • RLlib • Tianshou
Optimization & Config: Optuna • Ray Tune • Pydantic • SpaX
Data: NumPy • Pandas • Numba • Apache Spark • MongoDB • InfluxDB
Deployment: FastAPI • Docker • Linux • CI/CD

📍 Based in Iran, actively seeking opportunities in research labs and tech companies where pushing boundaries is the baseline expectation.