Thoughts on AI, insurance, and building systems
A staff-level deep dive into building distributed search infrastructure at scale. Covers query understanding, multi-stage retrieval, indexing strategies (inverted index, geo-spatial, vector search), ranking algorithms, and scaling to millions of QPS with sub-100ms latency.
How retrieval-augmented generation enables intelligent insurance chatbots and automation despite limited training data, strict regulations, and the need for verifiable, compliant responses in a risk-averse industry.
A practical introduction to designing machine learning systems—from data pipelines and model serving to monitoring, scaling, and handling feedback loops. Essential patterns and trade-offs every ML engineer should understand.
Why you shouldn't attempt to solve problems when your thinking mechanisms are shot. Like a machine learning algorithm needs training data to recognize patterns, your brain needs exposure to solutions before it can generate good ones. Stop speculating with zero training data—consume solutions first, solve later.
How insurers can move from manual rate filings to automated, data-driven pipelines that balance scale, accuracy, and regulatory compliance. Design principles for blending statistical rigor with flexible machine learning to enable transparent, regulator-friendly dynamic pricing.
How collaborative filtering and content-based methods can personalize coverage recommendations, optimize cross-sell strategies, and improve customer retention while respecting regulatory constraints.
Approaches to craft risk signals that improve predictive power while staying explainable to actuaries, regulators, and underwriters.
Understanding the metrics that matter for model evaluation in actuarial science—and why AUC alone isn't enough.