rhoai-autorag

name	rhoai-autorag
metadata	{"author":"rhoai3-demo","version":"1.0.0","platform-family":"rhoai","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"RHOAI Platform"}
description	Use when documenting, reviewing, or operating Red Hat OpenShift AI AutoRAG workflows from the official Working with AutoRAG guide: Technology Preview posture, RAG pattern optimization, JSON ground-truth evaluation data, English document and file-format limits, Llama Stack and remote Milvus prerequisites, Gen AI studio AutoRAG dashboard runs, externally imported AutoRAG pipeline naming, leaderboard evaluation, faithfulness/correctness/ context-correctness metrics, generated indexing and inference notebooks, search-space defaults, and recommended embedding model guidance. Do NOT use for Llama Stack server/provider/vector-store configuration (use rhoai-llama-stack), generic Gen AI playground workflows (use rhoai-gen-ai-playground), AI Pipelines server administration (use rhoai-ai-pipelines), S3 object operations outside AutoRAG (use rhoai-s3-object-storage-data), workbench/IDE workflows outside generated AutoRAG notebooks (use rhoai-data-science-ide-workflows), or live cluster changes without the OpenShift safety guard.

RHOAI AutoRAG

Use this skill for Red Hat OpenShift AI AutoRAG user workflows on the active product baseline in docs/PLATFORM_BASELINE.md.

Source Grounding

Read references/source-capture.md before using product workflow details. Official Red Hat documentation is product authority. This skill adapts the official Working with AutoRAG guide to this repo's demo workflow and governance review model.

Scope

This skill covers:

• AutoRAG Technology Preview support posture
• AutoRAG concept model, RAG patterns, and search space terminology
• English-only document support and Technology Preview limitations
• JSON evaluation data with questions, expected answers, and source document IDs
• Gen AI studio AutoRAG optimization run creation
• Llama Stack connection, foundation model, embedding model, and remote Milvus prerequisites
• document source rules for S3-compatible storage and local upload
• externally imported AutoRAG pipeline naming rules
• leaderboard review and pattern detail interpretation
• faithfulness, answer correctness, and context correctness metrics
• generated indexing and inference notebook handoff to a workbench
• user-configurable parameters and non-configurable search-space defaults
• recommended embedding model guidance from the official docs

Use other skills for adjacent work:

• rhoai-llama-stack for Llama Stack Operator activation, LlamaStackDistribution, providers, vector stores, Milvus registration, model availability, API behavior, OAuth/ABAC, CA trust, and HA/autoscaling
• rhoai-gen-ai-playground for product playground RAG uploads, prompt management, MCP testing, custom endpoints, and AI asset endpoint workflows
• rhoai-ai-pipelines for pipeline server configuration, imported pipeline management, pipeline run lifecycle, logs, and DSPA troubleshooting
• rhoai-s3-object-storage-data for object storage operations from workbenches
• rhoai-project-workflows for projects, workbenches, project connections, access, and cluster storage
• rhoai-data-science-ide-workflows for running generated indexing and inference notebooks in a workbench
• rhoai-model-evaluation for formal evaluation evidence outside the product AutoRAG leaderboard

Demo Policy

For this repo:

• Label AutoRAG as Technology Preview in READMEs, runbooks, presentations, and demo scripts. Do not present it as production SLA-backed.
• Treat AutoRAG as a product workflow for exploring and selecting RAG configurations, not as a substitute for governed application implementation.
• Require Gen AI studio, a data science project, Llama Stack connection, and remote Milvus before promising an AutoRAG run.
• Do not claim inline Milvus support for AutoRAG. The official AutoRAG guide requires remote Milvus.
• Use English-language documents for AutoRAG demos unless the active official baseline changes.
• Keep documents in supported formats: PDF, DOCX, PPTX, Markdown, HTML, or TXT.
• Do not claim OCR, embedded-image processing, or PDF table-structure detection during Technology Preview.
• Keep at most three foundation models and two embedding models selected for a run.
• Treat optimization runs as immutable after creation. Use AI Pipelines run workflows to stop, archive, or delete the underlying pipeline run.
• Use generated indexing and inference notebooks as reviewable handoff artifacts. Do not treat them as the final production RAG application.
• Use leaderboard scores together with sample Q&A review. Do not select a pattern solely because one metric is high.

Workflow

1. Confirm the active baseline in docs/PLATFORM_BASELINE.md.
2. Read references/source-capture.md and references/official-doc-extraction.md.
3. Confirm Technology Preview posture is acceptable for the requested demo surface.
4. Decide whether the task is:
   • AutoRAG concept or README authoring
   • prerequisite review for Gen AI studio, Llama Stack, Milvus, S3, and workbench access
   • JSON evaluation data design
   • optimization run design
   • external AutoRAG pipeline import review
   • leaderboard and pattern selection
   • indexing or inference notebook handoff
   • metric and search-space interpretation
5. Use examples/autorag-patterns.md for focused review patterns.
6. For live cluster work, follow the OpenShift safety guard in AGENTS.md.
7. Validate with references/validation-checklist.md.

References

• references/source-capture.md
• references/official-doc-extraction.md
• references/validation-checklist.md
• examples/autorag-patterns.md