About me
I build AI systems for problems where the data is messy and the answer actually matters.
I’m a senior data scientist with roots in cybersecurity. I care less about a flashy demo and more about what happens after it meets real data, real constraints, and the people who have to use the result.
My work has moved between security and enterprise learning, with machine learning and production engineering connecting the two. That has included phishing detection, threat intelligence, enterprise RAG, semantic search, and AI products for learning and knowledge access.
How I got here
At Cisco Talos, I worked with large security datasets, threat-intelligence databases, reporting automation, and phishing detection. I saw analysts face the same classification problems over and over, and started using machine learning to make that work more consistent and scalable.
At Cisco Learning & Development, I led a generative AI services team and helped build a production RAG assistant on AWS for Cisco U. The work included semantic search, knowledge retrieval, telemetry, and user feedback—the parts that turn a promising model into a useful learning product.
At ThreatSTOP, I work on machine-learning pipelines for phishing detection in DNS traffic. I’m also exploring AI agents that can gather evidence, explain their reasoning, and be honest about what they do not know—an approach that applies well beyond security.
How I work
Missing fields, stale data, changing behavior, API limits, and false positives are part of the design—not cleanup for later.
A useful system should show its evidence, its limits, and why it reached a conclusion. Confidence is context, not decoration.
A technically correct model is not enough. The output has to help a person decide what to do next.
What I like building
I’m happiest when I can follow a problem all the way from raw data to a decision someone can use. That might be a security classifier with layered context, a dependable recurring pipeline, or a knowledge product that gives a grounded answer instead of a confident guess.
Right now, I’m especially interested in production ML architecture, retrieval quality, useful AI agents, enterprise knowledge systems, and the security problems where those ideas are especially valuable.
For me, production-ready also means planning for security and privacy, scale, data quality, evaluation, logging and observability, cost, latency, failure recovery, versioning, and rollback. A model is only one part of the system that has to keep working.
Skills overview
Want the short version?
For a quick view of my roles, experience, and technical background, start there.
RAG-based LLM agent