Project · Write a fine-tuning go/no-go memo

Take a real project (yours or a public case study) and produce a one-page go/no-go memo: the decision (fine-tune / RAG / prompt), the target metric, the baseline you'd measure first, the data scope you could actually build, and the strongest counterargument. The memo must survive a skeptical engineer's review.

• A one-page go/no-go memo with an explicit fine-tune / RAG / prompt recommendation.
• A named target metric plus the baseline-with-best-prompt you'd measure before training.
• A realistic dataset scope (count + source) and a no-regression guardrail.

• The recommendation follows the decision tree and the reasoning is explicit, not asserted.
• The metric is measurable and a concrete baseline-measurement plan is stated.
• Data scope is realistic (buildable in days, not a fantasy million examples).