A Python architecture testing framework inspired by ArchUnit for Java, designed to help developers enforce clean architecture principles, dependency rules, and maintainable code structures in Python projects. As a contributor to the project, I have been involved in extending functionality, improving the framework, and supporting the long-term development of the library as it continues to grow within the Python ecosystem.
Repo: ArchUnitPython
Technologies: Python, Software Architecture, Testing Frameworks, Static Analysis, Clean Architecture
Highlights:
- Contributes to an open-source architecture testing framework inspired by the widely adopted Java ArchUnit library.
- Helps developers define and automatically validate architectural constraints within Python codebases.
- Focuses on maintainability, modularity, and scalable software design principles.
- Active contributor with increasing ownership and involvement in the future direction of the project.
A reinforcement learning approach to optimize food delivery logistics, addressing the Restaurant Meal Delivery Problem through an enhanced Anticipatory Customer Assignment framework. The project introduces RL-ACA, a novel algorithm that uses dynamic postponement strategies learned through Deep Q-Networks to optimize delivery assignment and bundling decisions. The system is comprehensively validated using real-world Meituan data (647,395 orders across 22 districts) and features statistical analysis across multiple operational contexts.
Repo: RMDP_Algorithm Technologies: Python, PyTorch (Deep Q-Network), NumPy, Pandas, Statistical Analysis, Real-time
Highlights:
- Achieves a 5.5% reduction in average distance per order and 1.5 percentage point lower idle rates through intelligent postponement decisions, improving driver efficiency and platform sustainability.
- Demonstrates superior performance in high-stress scenarios with 4.4 percentage point advantage in on-time delivery rates, showcasing adaptability under operational pressure.
- Validates performance across 120 real-world scenarios with statistical significance testing, providing robust evidence of algorithm effectiveness in diverse urban delivery contexts.
- Features comprehensive benchmarking framework comparing RL-ACA against baseline methods, with detailed analysis of trade-offs between routing efficiency and delivery timeliness across stakeholder priorities.
A reinforcement learning framework for modeling returns and decision-making in omnichannel retail. The project leverages a Hierarchical Markov Decision Process (HMDP) and the Proximal Policy Optimization (PPO) algorithm to optimize ordering and allocation strategies for retailers operating online and offline channels with resellable returns.
Repo: OmniChannel-RL Technologies: Python, TensorFlow, Keras, NumPy, Gym Highlights: Achieved a 3% reduction in total costs and up to a 17% increase in service levels. The model is based on the framework outlined in the paper J. Goedhart, R. Haijema, and R. Akkerman (2023), showcasing the potential of reinforcement learning in complex, hierarchical decision-making environments.
A reinforcement learning solution for the Supply Chain Beer Game. Modeled the game's supply chain dynamics in Python (NumPy, Pandas) and implemented Q-Learning to optimize ordering strategies. Achieved a 31% reduction in total costs by modifying the state space, demonstrating the potential of RL in supply chain optimization.
Repo: Beer-Game-RL Technologies: Python, NumPy, Pandas, Q-Learning Highlights: Combines supply chain simulation with reinforcement learning to explore automated decision-making and cost minimization in complex systems.
Languages
Visualization
