MOC: AI & Prompts

This chapter is organized around the jobs engineers and analysts actually need to do with AI in production: design prompts that behave consistently, build LLM pipelines that stay observable and governable, decide where AI belongs in data engineering, and apply it safely in ESG and index workflows.

mindmap
  ((Prompt Craft))
    (Foundations)
    (Architecture)
    (Applied prompting)
    (Model-specific tuning)
    (Debugging)

Prompt Craft

domain-prompt-craft

Prompt fundamentals, prompt architecture, applied prompting, model-specific tuning, and systematic prompt debugging.

mindmap
  ((AI Platform Patterns))
    (Pipeline architecture)
    (Retrieval and tool use)
    (Tooling landscape)
    (Evaluation)
    (Observability)
    (Governance)

AI Platform Patterns

domain-ai-platform-patterns

LLM pipeline architecture, retrieval and tool use, current tooling categories, evaluation, observability, and operational guardrails.

mindmap
  ((AI in Data Engineering))
    (Ingestion and extraction)
    (Normalization and reconciliation)
    (Catalog and lineage support)
    (Testing and incident support)
    (Guarded developer productivity)

AI in Data Engineering

domain-ai-in-data-engineering

Where AI helps across ingestion, extraction, reconciliation, cataloging, incident support, documentation, and developer productivity in production data platforms.

mindmap
  ((AI in Financial Workflows))
    (ESG analytics)
    (Taxonomy mapping)
    (Controversy triage)
    (Index methodology support)
    (Corporate actions QA)
    (Approval boundaries)

AI in Financial Workflows

domain-ai-in-financial-workflows

ESG document intelligence, issuer-level enrichment, methodology interpretation, index maintenance support, and governance boundaries for regulated operations.

AI and Prompts Cross-References

• moc-data-architecture for idempotent pipelines, data contracts, retries, and validation patterns that AI layers must inherit rather than bypass.
• moc-observability for tracing, incident response, auditability, and alerting patterns extended here to AI-specific telemetry.
• moc-dataops for release discipline, rollback criteria, and approval workflows that constrain model changes in production.
• moc-gcp for service boundaries, secret management, cost control, and deployment patterns relevant to hosted AI systems.
• moc-github-actions for regression testing, CI enforcement, and review automation around prompts, eval suites, and model configurations.
• moc-financial-domain for the benchmark, ESG, and index context that makes these AI patterns operationally meaningful rather than generic.