About

Reinforcement learning, optimization and forecasting — mostly in production

I would love to solve practical problems using data-driven approaches. I try my best not to confine myself to certain specific yet limited tools and solutions and rather use the best suitable solution for the problem in hand but through my data science and machine learning career, I mostly worked on decision-making under uncertainty type of applications. In practice that means reinforcement learning, mathematical optimization and forecasting — usually applied to systems where a wrong decision costs real money, and where the textbook version of the method turns out to need considerable adjustment before it survives contact with production.

If you’re thinking about working together, how I work covers the practical side — what I’m useful for, what I’m not, and how to get the most out of a conversation.

What I’m working on

I’m a Senior Data Scientist at Enverus, where I own the end-to-end development and operation of locational marginal price forecasting across thousands of nodes in electricity markets — feature engineering, applied research, distributed training, deployment, APIs, monitoring and the unglamorous maintenance that keeps a forecast trustworthy.

Before Enverus I was a machine learning engineer at Bits in Glass, in which I contributed building a hierarchical RAG platform over 20 million oil-and-gas documents and a dynamic fuel-pricing system that combined integer programming with contextual bandits. Earlier still I built RL-based surge pricing at Snapp, the largest ride-hailing platform in the Middle East, running across roughly 200,000 orders a day. My CV has the full history.

Research

I have PhD Geomatics Engineering at the University of Calgary, and BSc and MSc in Industrial Engineering from top universities in Iran (Amirkabir University of Technology and IUST). My research applies hierarchical reinforcement learning to real-time policy optimization — vehicle routing, disaster-response logistics and transportation problems where demand is stochastic, the fleet is heterogeneous, and a decision has to be made now rather than optimally.

Published work appears in Computers & Operations Research, Transportation Research Part C, ACM SIGSPATIAL and the KDD supply-chain workshop. The research page lists everything.

The book

I’m the author of Applied Reinforcement Learning (Manning), currently in early access with full release expected January 2027. It covers RL for business optimization — warehouse logistics, dynamic pricing, routing, resource allocation — through to fine-tuning language models with PPO, GRPO and verifiable rewards.

I wrote it because most RL material stops at games and simulations. The gap between a working policy in a clean environment and one that holds up against real constraints, messy data and stakeholders who need to understand the decision is where the actual difficulty lives, and it deserved a book.

This site is where the shorter-form version of that lives: articles and tutorials, notes on papers I’m reading, implementations, and occasional writing about the professional side of doing this work. Some of it is paired with videos on my YouTube channel, PhiloAI.

Elsewhere

I’m always glad to hear from people working on RL or optimization in production, and from students trying to get started. If you’re writing about something in this space, send it to me — I’d rather read your work than another press release.

Before a first project, how I work is worth five minutes.