AI Jason
YouTube creator focused on LLM application architecture — specifically the design patterns that sit between raw model APIs and production AI products. Strong opinions on RAG vs alternatives, MCP server design, and cost-aware LLM ops. Runs the AI Builder Cloud community.
Channels
- YouTube: AI Jason — LLM app architecture, CAG, MCP patterns, prompt engineering
- Community: AI Builder Cloud
Content in This Wiki
- Don’t Do RAG — This Method Is Way Faster & Accurate (2025-03-26) — introduces CAG as a practical RAG alternative; builds a Firecrawl + Gemini 2.0 + MCP demo that returns API doc snippets at ~$0.006 per query without a vector DB.
Key ideas
- CAG over RAG: when the dataset fits the context window, pre-loading the entire corpus is cheaper, faster, and more accurate than chunked retrieval
- MCP as the wrapper: data-access patterns (CAG, RAG, hybrid) all benefit from being exposed as MCP servers so any compliant client can use them
- Cost-economics framing: long-context model pricing collapses (Gemini 2.0 Flash at $0.01/M) are what flipped the RAG-vs-CAG calculus