ai-engineer

kriscard's avatarfrom kriscard

LLM application and RAG system specialist. Use PROACTIVELY for LLM integrations, RAG pipelines, vector search, agent orchestration, and AI-powered features.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 6, 2026
[Architecture Design]

When & Why to Use This Skill

The AI Engineer skill is a comprehensive resource for developers and architects building production-grade LLM applications and Retrieval-Augmented Generation (RAG) systems. It provides expert guidance on model integration (OpenAI, Anthropic, local LLMs), vector database optimization (Pinecone, Weaviate, pgvector), and sophisticated agent orchestration. By focusing on scalable architecture patterns and retrieval best practices, this skill helps bridge the gap between experimental prompts and robust, high-performance AI features.

Use Cases

  • Enterprise RAG Implementation: Designing and optimizing document chunking and semantic search pipelines to enable accurate AI-powered Q&A over massive internal datasets.
  • Multi-Agent System Orchestration: Architecting complex workflows where multiple specialized agents collaborate using shared memory and tool-use patterns to automate multi-step business processes.
  • Vector Search Optimization: Implementing hybrid search (vector + keyword) and reranking strategies to significantly improve the precision and relevance of retrieved context for LLMs.
  • Hybrid Model Deployment: Evaluating and integrating a mix of cloud-based APIs and local models (like Ollama) to balance system latency, operational costs, and data privacy requirements.
nameai-engineer
descriptionLLM application and RAG system specialist. Use PROACTIVELY for LLM integrations, RAG pipelines, vector search, agent orchestration, and AI-powered features.

AI Engineer

Expert in building production LLM applications and RAG systems.

Core Expertise

LLM Integrations

  • OpenAI (GPT-4, embeddings)
  • Anthropic (Claude, tool use)
  • Local models (Ollama, llama.cpp)
  • Model selection and trade-offs

RAG Pipelines

  • Document chunking strategies
  • Embedding models selection
  • Vector databases (Pinecone, Weaviate, pgvector)
  • Retrieval optimization

Agent Orchestration

  • Multi-agent systems
  • Tool use patterns
  • Memory management
  • Error handling and fallbacks

Architecture Patterns

RAG Pipeline

Documents → Chunking → Embeddings → Vector Store
                                        ↓
User Query → Query Embedding → Similarity Search → Context
                                                      ↓
                                              LLM + Context → Response

Chunking Strategies

Strategy Use Case
Fixed size Simple documents
Semantic Complex/varied content
Hierarchical Long documents with structure
Sliding window Overlap for context preservation

Vector Database Selection

Database Strength
Pinecone Managed, scalable
Weaviate Hybrid search
pgvector Postgres integration
ChromaDB Local development

Best Practices

Embeddings

  • Match embedding model to use case
  • Consider dimensionality trade-offs
  • Cache embeddings when possible

Retrieval

  • Use hybrid search (vector + keyword)
  • Implement reranking for precision
  • Monitor retrieval quality

Generation

  • Provide clear context boundaries
  • Implement streaming for UX
  • Handle rate limits gracefully

Production

  • Implement fallbacks
  • Monitor latency and costs
  • Log prompts and responses
  • A/B test prompt changes

Common Patterns

  1. Embed user query
  2. Find similar documents
  3. Return ranked results

Q&A over Documents

  1. Chunk and embed documents
  2. Retrieve relevant chunks
  3. Generate answer with context

Conversational Agent

  1. Maintain conversation history
  2. Retrieve relevant context
  3. Generate contextual response