Posts tagged #LLM

Designing LLM Evals That Catch Real Regressions

LLM features break in ways normal unit tests miss. A useful eval suite starts with real examples, checks behavior instead of vibes, and separates model quality from product quality.

RAG in 60 Seconds

An LLM only knows what it was trained on. RAG lets it go look something up before it answers. Here's the whole loop in six steps.

Streaming LLM Responses in Next.js Without a Framework

Token-by-token streaming makes an AI feature feel instant. Here's how I stream model output straight from a Next.js Route Handler using the Web Streams API — no extra SDK, no client library, about 40 lines total.

Getting Reliable JSON Out of an LLM

"Respond in JSON" in the prompt is not a contract — it's a suggestion the model breaks under pressure. Here's how I get schema-valid structured data every time using tool calling, schema validation, and a tight repair loop.