TayagornManorat

A business-critical report had accumulated transformation and reporting logic that made refreshes slow and future changes harder to maintain.

I separated responsibilities across the stack: reusable transformations in SQL views, governed relationships in the semantic model, and responsive user logic in DAX. Validation checks reconciled results across enterprise systems before release.

Refresh time fell from 60 minutes to 3, while role-based access and metric switching gave different audiences one maintainable self-service reporting experience.

My health data lived across disconnected platforms, which made it difficult to understand recovery, nutrition, body composition, and long-term goals together.

I built ingestion pipelines for WHOOP, FatSecret, and Apple Health, handled multiple OAuth flows, modeled the data in a 20+ table SQLite schema, and exposed it through an MCP server with more than 16 purpose-built tools.

The system now gives an LLM structured access to more than 1.5 million records while Grafana provides a separate visual layer for trends, validation, and day-to-day monitoring.

Temperature contracts on separate prediction markets can disagree with each other and with the probability implied by forecast data and historical climatology.

I built a polling and normalization layer, compared market-implied probabilities against a weather model, and isolated each paper strategy with its own bankroll, decision history, and Kelly-based position sizing.

The result is an observable research environment where strategies can be compared without mixing capital, market signals, or performance histories.