Prompt Craft
Prompt craft is the discipline of turning a vague AI request into an explicit operating contract: what the model should do, what evidence it may use, what output shape it must respect, and how the result will be evaluated.
mindmap ((Prompt Craft)) (Foundations) (Architecture) (Applied prompting) (Model-specific tuning) (Debugging and versioning)
Core definitions for prompts, tokens, context, grounding, hallucinations, and instruction hierarchy, with finance-aware examples from ESG and index operations.
How to design prompt templates, delimiters, schema-bound outputs, decomposition patterns, and reusable prompt contracts for production systems.
Prompt patterns for research, extraction, summarization, code review, SQL generation, and domain-specific workflows in data engineering and financial operations.
How prompting changes across frontier hosted models, long-context models, reasoning-oriented models, and smaller local or task-specialized models.
How to diagnose bad outputs, build regression suites, version prompt changes, and decide whether the problem is the prompt, the model, the retrieval layer, or the surrounding application.