ai-engineering-patterns

name: ai-engineering-patterns description: AI engineering patterns relevant to GGV3 - agentic RAG, financial analysis, MCP infrastructure, context engineering, and multi-agent deep research. Source patterns from patchy631/ai-engineering-hub. trigger: RAG implementation, AI search, financial analysis, MCP server, context engineering, knowledge graph, agent memory, deep research

AI Engineering Patterns for GGV3

Curated patterns from patchy631/ai-engineering-hub relevant to the GearGrind tactical marketplace.

Agentic RAG (Retrieval-Augmented Generation)

When to use

• Product search & discovery with semantic understanding
• Customer support with document-aware responses
• Content moderation with context-aware analysis

Pattern

1. Document Ingestion → Chunk + embed documents
2. Query Processing → Analyze intent + generate search queries
3. Retrieval → Vector search + keyword search (hybrid)
4. Agent Decision → If retrieval insufficient, fall back to web search
5. Response Generation → Synthesize answer from retrieved context

Key Implementation Details

• Use embedding models (e.g., OpenAI ada-002, Cohere embed) for vector representation
• Hybrid search: combine vector similarity with BM25 keyword matching
• Implement re-ranking to improve retrieval relevance
• Add guardrails for hallucination prevention

Financial Analysis Patterns

When to use

• Auction pricing intelligence
• Market trend analysis for tactical gear
• Revenue forecasting and reporting

Pattern

• LLM-powered analysis with structured output (TypeScript interfaces)
• MCP server for financial data access
• Chart generation from computed metrics

Context Engineering

When to use

• Long-running agent sessions
• Multi-step complex tasks
• Knowledge persistence across conversations

Pattern

1. Context Collection → Gather relevant docs, conversation history, tool outputs
2. Context Compression → Summarize, prioritize, trim to fit window
3. Memory Persistence → Store important findings in knowledge graph (Graphiti/Zep)
4. Context Injection → Load relevant context at each step

MCP Infrastructure

When to use

• Connecting to external data sources (DB, APIs, services)
• Providing tools to AI agents
• Unified interface for heterogeneous data

Pattern

• Define tools with typed schemas
• Implement resource providers for data access
• Use server-sent events for streaming responses
• MindsDB pattern: unified MCP for all data sources

Multi-Agent Deep Research

When to use

• Complex investigation tasks
• Market research for tactical gear categories
• Competitive analysis

Pattern

1. Query Decomposition → Break complex question into sub-queries
2. Parallel Research → Multiple agents research different aspects
3. Source Evaluation → Verify and rank sources by reliability
4. Synthesis → Combine findings into structured report
5. Iteration → Identify gaps, generate follow-up queries