Software Engineering Manager / AI & Security Systems
I lead engineering teams across skills and backgrounds, from security to AI/ML to software and cloud infrastructure, at large technology companies. These days the job is less about the code than the conditions around it: setting direction across teams, growing the engineers and leads who will set it next, and keeping the work anchored to real customer problems, so the AI and security platforms we ship hold up at scale, under adversarial pressure, and keep delivering value to customers long after launch.
I'm a software engineering manager working at the intersection of artificial intelligence and security. I've spent my career in large technology companies, leading engineers of every stripe, from security to AI/ML to software and cloud infrastructure, as they build and operate the platforms our customers depend on.
A model is the easy part. What takes real engineering is everything around it: the data pipelines, the services, the infrastructure underneath, the way it degrades, and the rotation that catches it when it does. Most of those calls belong to the engineers closest to the problem. My job is making sure they have the context, the cover and the room to make them well, and that what we ship still holds up when the conditions change.
I also teach graduate-level computer science and data courses at a university near me. Explaining this work to people who will push back on it keeps the fundamentals sharp and makes me a better manager of the engineers doing it. Role history and specifics live on my LinkedIn rather than here.
Supporting the people who build these systems — and staying close enough to the engineering to be useful to them.
Hiring well, then getting out of the way: clear context, honest feedback, and the stretch assignment someone is nearly ready for. Removing the dependencies, decisions and ambiguity that stall good engineers, and making sure the people doing the work get the credit for it.
The systems around the models: data pipelines, feature and serving infrastructure, evaluation and rollout, cloud footprint and cost. Helping the team turn promising prototypes into services other teams can depend on, with clear contracts and a sane operational story.
Detection and response platforms that protect customers from fraud, abuse and anomalous activity, built to keep working while the adversary adapts. Layered signals, fast iteration on rules and models, measurable precision and recall, and humans in the loop where the stakes demand it.
Telemetry, forecasting and observability so problems surface before customers find them. Capacity and cost planning, incident response, and the unglamorous work of making a system boring to operate.
The fastest way to find out whether you understand something is to teach it to people who will ask why.
I teach graduate-level computer science and data courses at a university near me — working engineers and career-changers, mostly, which keeps the material honest and applied rather than theoretical.
Presented at the Network Traffic Measurement and Analysis (TMA) Conference on the practical limits of learning models over network telemetry.
Free lessons and walkthroughs on YouTube, notes on engineering and machine learning on Medium, and code in the open on GitHub. Most of what I know came from people who published it for nothing — this is the interest payment.
Earlier research on adversarial behaviour online, fraud detection over relationship graphs, and the practical limits of running learning models in operational systems.
A data enrichment scheme that surfaces signal hidden in the relationships between entities, plus an efficient method for keeping those social-network connections up to date.
Measuring the quality of incentivized reviews on Amazon.com, how far they deviate from organic reviewer behaviour, and the effect they have on product ratings.
Forecasting and characterising network traffic time series with statistical (SARIMA) and deep learning (LSTM) techniques, and examining what it really costs — in data, training and retraining — to keep those models useful in an operational network.
A multi-resolution analysis of the most influential slice of Twitter: how the elite subgraph is structured, which communities it splits into, and how those connections shift over time.
A longitudinal look at incentivized reviews: how review campaigns form and evolve, and how they distort the rating signal buyers rely on.
The projects that started as a weekend curiosity and refused to stay one.
Built in 2009 during my B.Sc. as the first peer-to-peer question-and-answer website in Persian. It grew into one of the most popular Persian Q&A communities and now lives on as soja.ai.
Web presence for academic conferences including PAM 2020, ICEE 2012 and IPG 2012–13, plus a full conference-management system in Persian that ran submissions and reviewing for several conferences between 2012 and 2014.
Long exposures, city nights and whatever the Pacific Northwest is doing with the light that day. A useful reminder that not everything worth doing needs a metric attached to it.
A few frames from the Pacific Northwest and beyond. More on Instagram.
Building an AI or security platform, growing an engineering team, or wrestling with a detection system that keeps getting evaded? Happy to compare notes. My inbox is open.
Address is assembled in your browser — sorry, scrapers.