Bob

I have 50k labeled support tickets to route into 12 categories. Prompting GPT 5 works but costs add up at volume. When does fine tuning a small model pay off?

I need guaranteed parseable JSON. Response format json object helps but it still occasionally wraps things in prose or markdown fences. What is the current best practice?

Building retrieval over long technical docs. Fixed 512 token chunks are simple but cut mid sentence and hurt recall. Is semantic chunking worth the complexity, or is overlap enough…