ai-engineer_skill

Overview

This skill helps engineers design and deploy production-grade LLM applications, retrieval-augmented generation (RAG) pipelines, and multi-agent orchestration. It focuses on practical choices: model selection, embedding strategies, vector stores, and operational trade-offs. Use it when you need repeatable patterns and hardened practices for LLM systems rather than general backend work.

How this skill works

The skill inspects architectural choices and prescribes concrete pipelines: document chunking, embedding selection, vector database layout, and retrieval strategies. It lays out agent orchestration patterns, memory handling, tool-use flows, and production hardening steps like logging, fallbacks, and monitoring. Outputs include recommended components, trade-offs, and step-by-step integration guidance.

When to use it

• Building RAG pipelines to answer questions from documents
• Choosing embeddings and vector stores for semantic search
• Integrating managed or local LLMs into applications
• Designing multi-agent systems and tool-use orchestration
• Optimizing retrieval, reranking, and generation for accuracy

Best practices

• Match embedding model and dimensionality to your retrieval needs and cost constraints
• Chunk documents using semantic or hierarchical strategies for complex sources
• Combine vector similarity with keyword filters (hybrid search) and reranking for precision
• Cache embeddings and monitor retrieval quality to reduce latency and cost
• Implement streaming responses, rate-limit handling, and graceful fallbacks in production

Example use cases

• Customer support knowledge base with semantic search and RAG answers
• Internal document assistant: chunk, embed, and query policy or legal corpora
• Agent system that chains tools, maintains memory, and falls back to safe defaults
• Local development with ChromaDB or pgvector before migrating to managed stores like Pinecone
• A/B testing prompt templates and retrieval parameters to optimize answer quality