القائمة
Serve, scale, and safely roll out a RAG pipeline with health checks, caching, and canary releases.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use this skill when building, debugging, or improving Retrieval-Augmented Generation systems, including chunking, vector database selection, hybrid search, reranking, multimodal RAG, code documentation RAG, retrieval latency, and production RAG architecture.
Chunk nested documents into parent-child levels so retrieval can move from broad sections to fine-grained passages.
Use semantic boundaries and embedding similarity to chunk text for higher-relevance retrieval.
Route RAG chunking decisions across semantic, hierarchical, sliding-window, contextual-header, and framework-selection strategies.
Use overlapping windows to preserve context across chunk boundaries while controlling retrieval size.
Reduce retrieval latency with caching, batching, and index-level optimization.
| name | production-rag-deployment |
| title | Production RAG Deployment |
| description | Serve, scale, and safely roll out a RAG pipeline with health checks, caching, and canary releases. |
| allowed-tools | ["Read","Grep","Glob","Bash"] |
| category | deployment |
| tags | ["serving","scaling","canary","observability"] |
This skill covers turning a working RAG pipeline into a reliable production service: exposing it behind an API, scaling retrieval and generation independently, caching hot paths, and rolling out changes without breaking users.
Prototype RAG code assumes one user, a warm index, and unlimited time. In production it faces concurrency, cold starts, provider timeouts, and cost limits. Without serving discipline, latency spikes, failures cascade, and a bad deploy hits every user at once.
Wrap retrieval and generation in an API with timeouts, retries, and health checks for each dependency.
Why: Explicit timeouts and health checks stop a slow vector store or LLM from stalling the whole request path.
Scale the retrieval and generation tiers independently and cache embeddings and frequent answers.
Why: Retrieval and generation bottleneck differently, and caching removes repeated cost and latency on common queries.
Release changes to a small traffic slice first and define a degraded path when a component is unavailable.
Why: Canary releases and fallbacks contain the blast radius of a bad deploy or a failing dependency.
For infrastructure patterns, see the Qdrant Cloud documentation and the Ray Serve LLM guide.
