ml-pipeline-creation

# ML Pipeline Creation Build reproducible ML workflows whose inputs, outputs, lineage, and promotion criteria are explicit. Prefer the project's existing orchestrator and conventions; do not introduce a platform merely to demonstrate one. ## Required Inputs - Business objective and measurable model acceptance criteria - Data sources, ownership, sensitivity, and expected refresh cadence - Existing preprocessing, training, evaluation, and serving code - Target environments and available orchestration or CI system - Compute, cost, latency, reproducibility, and compliance constraints If critical details are missing, state assumptions and design a platform-neutral pipeline before selecting an implementation. ## Output Contract Produce: 1. A dependency graph of pipeline stages and artifacts 2. A versioned pipeline definition or implementation 3. Explicit schemas for every stage input and output 4. Data, model, and environment versioning rules 5. Evaluation and promotion gates with failure behavior 6. Observability, retry, backfill, and rollback procedures 7. A verification record showing how the pipeline was tested ## Workflow 1. **Inspect the environment.** Identify the repository language, dependency manager, existing orchestration system, model framework, artifact store, and deployment path. Reuse established tools where possible. 2. **Define the contract.** Record the objective, data snapshot rules, target metric, baseline, acceptance threshold, resource budget, and de...