TECHNIQUE
Model Adaptation
Observed prompt-engineering-at-scale practice is sparse in this pool: Dropbox reports prompt optimization with DSPy, while LinkedIn reports task-specific prompt templates for semantic textual similarity in job search.
Use a prompt-optimization framework rather than only hand-written prompts.
1 of 2 operators with teardown evidence for prompt engineering at scale.Use prompt templates designed for a specific semantic task.
1 of 2 operators with teardown evidence for prompt engineering at scale.Embed prompt engineering inside retrieval/search workflows rather than presenting it as a standalone layer.
2 of 2 operators with teardown evidence for prompt engineering at scale.Both observed operators apply prompt engineering in search or retrieval-oriented products.
How prompt engineering is operationalized differs across the observed operators.
APPROACH 01
Prompt optimization via DSPy.
APPROACH 02
Task-specific prompt templates for semantic textual similarity.
| Name | Kind | When | Maturity |
|---|---|---|---|
| Versioned prompt registry | pattern | prompts ship like code: ids, diffs, owners, and rollback | established |
| DSPy | library | prompts tuned programmatically against eval metrics, not by hand | emerging |