recsys-pipeline-architect

recsys-pipeline-architect is a spec-and-scaffold skill for designing composable recommendation, ranking, and feed pipelines. It encodes a six-stage pattern -- Source, Hydrator, Filter, Scorer, Selector, and SideEffect -- popularized by xAI's open-sourced For You algorithm, and is explicitly an independent MIT reimplementation of that pattern with no code copied from the original. The problem it solves is the recurring plumbing challenge of any system that must pick "the top K items for a (user, context)": social feeds, content CMSs, RAG rerankers, task prioritizers, notification triage, search reranking, and ad ranking.

When invoked, it walks the user through an eight-step workflow: clarify the use case, identify candidate sources (in-network and out-of-network), list required hydrations, order the filters cheap-before-expensive, design a scorer chain (primary ML, combiner, diversity, business rules), define the selector, wire fire-and-forget side effects, and generate a runnable scaffold in the user's stack (TypeScript, Go, or Python). It deliberately surfaces key trade-offs rather than defaulting silently -- single score versus multi-action prediction, isolated versus joint candidate scoring, and online versus offline execution -- and enforces hard rules such as never inventing benchmark numbers, attributing the pattern correctly, avoiding trademark or "For You" branding, ensuring the generated scaffold actually runs, and never letting side effects block the response.

It is aimed at engineers building or migrating ranking and feed systems who already have (or plan to wrap) an ML or LLM scoring function and need the surrounding architecture. It explicitly excludes model architecture work, pure ML training pipelines, and operating a deployed pipeline. This is an architecture and code-generation helper, grounded entirely in a documented design pattern.