Applied, Not Theoretical
I build AI systems that ship inside real products. My physics thesis involved building an agentic LLM framework that autonomously detected particle physics anomalies - including LLM fine-tuning and reimplementing weakly-supervised detection pipelines from scratch.
That same rigor goes into production work.
What I Build
- LLM Integration & Agents - Structured systems around LLMs. Tool orchestration, prompt architecture, context management, output validation, graceful degradation. Reliable automation, not demos.
- Predictive Analytics - Full pipeline: data ingestion, feature selection, model training, backtesting, API serving. Automated retraining with performance monitoring.
- Anomaly Detection - Statistical + ML hybrid approaches. Financial data, user behavior, system metrics - anything with structured data and a definition of "normal."
- Data Pipelines - Automated collection, transformation, and model feeding on schedule. Monitoring, alerting, self-healing.
My Approach
The question isn't "how do we add AI?" - it's "what problem are we solving, and is AI the right tool?"
When it is, I build it into your existing product. When it isn't, I'll tell you.
What I Work With
LLMs - OpenAI, Anthropic, open-source; RAG, function calling, agent frameworks Classical ML - XGBoost, scikit-learn, statistical modeling Infrastructure - Python, Node, scheduled pipelines, model versioning, API serving