pipeline▌

Audits Python + BigQuery pipelines for cost safety, idempotency, and production readiness with exact patch locations. \n \n Analyzes every BigQuery job trigger and external API call to identify cost exposure, loop-driven query multiplication, and missing maximum_bytes_billed limits \n Enforces dry-run and execute modes with explicit prod confirmation, partition filter validation, and scan-size optimization \n Validates idempotent writes using MERGE, staging tables, or dedup logic; flags unsafe a

Multi-step AI content creation pipelines combining image, video, audio, and text generation. \n \n Chains tools like FLUX (image generation), Wan 2.5 (animation), Kokoro TTS (voice), and OmniHuman (talking heads) into complete workflows via the inference.sh CLI \n Includes four ready-to-use pipeline templates: YouTube shorts (script → voiceover → background → animation → merge), talking head videos, product demos, and blog-to-video conversion \n Supports common patterns such as image-to-video-to

End-to-end MLOps pipeline orchestration from data ingestion through model deployment and monitoring. \n \n Covers five core pipeline stages: data preparation, model training, validation, deployment, and monitoring with DAG orchestration patterns (Airflow, Dagster, Kubeflow) \n Includes data validation, feature engineering, experiment tracking integration, and model versioning strategies across the full ML lifecycle \n Provides deployment automation patterns including canary releases, blue-green

Intelligent model routing, budget tracking, and retry logic to optimize LLM API costs without sacrificing quality. \n \n Routes requests to cheaper models (Haiku) for simple tasks and expensive models (Sonnet, Opus) only when complexity thresholds are met, reducing spend by 3–19x on routine work \n Tracks cumulative API costs with immutable dataclasses, enforces budget limits, and fails early to prevent overspend \n Implements narrow retry logic that retries only on transient errors (network, ra