A RAG app is only as good as the context it retrieves. For this demo, I built an ARC Raiders retrieval assistant that makes the retrieval layer visible instead of hiding it behind the final answer.…

Keyword search asks: "Did this text use the same words?" Semantic search asks: "Is this text about the same thing?" That shift is what makes embeddings so useful for AI apps, especially when users do…

I built a Next.js comparison tool for that: - Zero-shot prompt - Few-shot prompt - Chain-of-thought style prompt - AI SDK streamText() - useChat lanes for each pattern - Side-by-side outputs for the…

Prompting feels subjective until you compare patterns side by side. I built a Next.js prompt pattern lab that sends the same question through zero-shot, few-shot, and chain-of-thought system prompts…

I built a specification precision demo around that idea: Versioned system prompt library Zod schemas for structured output AI SDK generateText() endpoint with Output.object + Prompt specs for MCP…

Most AI bugs start before the model runs. I built a specification precision demo: - production-style system prompts - Zod schemas - AI SDK generateText() with Output.object and prompt specs for: tool…

I built an MCP joke agent inside my AI Engineer demo app with Vercel's AI SDK. The UI discovers tools from an HTTP MCP server at runtime, then a ToolLoopAgent uses an MCP client to call…