- Agent runtime, orchestration and microservices for an agentic CRM.
- Real-time voice agent and multimodal graph-RAG search.
The Hidden Layer
How AI systems actually work, on Medium and Substack.
I'm an AI architect and founder who came to software from physics. I'm joining Caterpillar as Senior Principal Architect for AI and ML, I lead Data & AI at fenel, and I founded MedBrain Technologies to bring agentic AI to clinical data. For over a decade I've shipped production ML in e-commerce and pharma, most recently leading 20+ engineers. Today I design agentic systems and the parts that make them trustworthy: evaluation, retrieval, and the data platforms underneath. My PhD at UBC modeled medical cyclotron targets at TRIUMF, and I still work like a modeler: start from first principles, then check the numbers. I also work where software meets hardware: at Aeris I built gas-sensing devices, their firmware and telemetry, and a cloud solver that pinpoints leaks, and I run local models and edge AI where there is no data center to lean on.
Multi-agent orchestrationMCPRAGGraph RAGVoice agentsKnowledge graphs
TensorFlowscikit-learnXGBoostCI/CDMonitoringW&B
AWSAzureKubernetesTerraformDagster
PythonSQLPySparkSnowflakeDuckDBIoT telemetry
A publication on how AI systems work, talks, and a patent from the forecasting work.
How AI systems actually work, on Medium and Substack.
A WIPO patent from my forecasting work at ODAIA.