A hash-based query cache that computes repeatable hashes of query inputs, persists results to MongoDB, applies TTL and scheduled refresh logic, and normalizes Oracle/MySQL numeric values to avoid string-corrupted aggregates. It is a self-contained pattern for building a resilient, reusable database-query cache layer in any Node.js backend.

A self-contained ARIMA forecasting service with automatic model configuration selection, validation splitting, and a linear regression fallback when ARIMA is not suitable. It demonstrates a pragmatic, production-oriented approach to univariate time-series prediction and could be extracted for use in other analytics applications.

A unified N-axis cross-table aggregation engine that replaces legacy pivot implementations. It uses sparse Y-tuple computation with a non-printable separator to avoid cartesian column explosions and includes preflight column-count checks to prevent browser crashes. This is a strong, reusable data-aggregation pattern for pivot/cross-table behavior in web applications.

A clean factory abstraction over multiple LLM providers (OpenAI, Anthropic, Bedrock, Qwen), exposing a common provider interface for chat and tool calling. It demonstrates a loosely coupled way to swap AI backends, which can be lifted into other projects needing multi-provider LLM integration.

A compact collection of reusable D3 formatting helpers used across the chart component suite, including number/axis formatting, label truncation, and measure-text-width calculations. It is a dependency-light, self-contained utility module that can be dropped into any D3/TypeScript visualization project.