Fine-Tuning
Additional training that adjusts a model's weights on your examples — powerful for narrow, high-volume tasks, and usually the wrong first tool for knowledge problems.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
Fine-tuning continues a model's training on your own examples so the behavior moves toward them: your format, your tone, your task. It changes the model itself, unlike prompting or retrieval, which change what the model sees.
Why it matters
Fine-tuning is regularly reached for too early. It does not reliably teach a model new facts — retrieval does that better, stays current, and keeps citations. Where fine-tuning earns its cost: a narrow task at high volume where a smaller tuned model can replace a larger general one, or a strict output style prompting can't hold.
How it works
Curate example pairs, train (full fine-tune or an adapter method like LoRA), then evaluate against the untuned baseline on a held-out set. The eval decides whether it shipped an improvement. Most enterprise problems resolve at the prompt, retrieval, or routing layer before a fine-tune is justified — which is why our default stack contains none.
Related resources
The pattern where an AI agent retrieves relevant context from your data before generating an answer — instead of relying only on what the model learned during training.
How an AI system decides which model to call for each step — based on privacy, cost, latency, quality, and what happens when a provider goes down.
Short for 'evaluations' — the test cases and harness that measure whether an AI workflow is working, before and after every change.
The total spend on language-model APIs across an organization — input tokens, output tokens, embeddings, fine-tuning — and the practice of attributing, optimizing, and budgeting it.