[Enhancement]: Develop SimBoard prototype for AI-assisted simulation summaries

[tomvothecoder]

Is your feature request related to a problem?

SimBoard users need faster ways to understand selected simulations and related diagnostics context. Today, relevant context may be spread across simulation metadata, diagnostics outputs, documentation, examples, and workflow notes. Users must manually connect these sources to answer common questions.

Describe the solution you'd like

Implement a narrowly scoped, read-only prototype in SimBoard that includes both:

• a FastAPI backend path for generating structured simulation summaries from authoritative SimBoard metadata, and
• a minimal SimBoard UI integration on the simulation detail page so users can request and read that summary in context.

Prototype intent:

• Feasibility prototype, not a production commitment and not a full shared E3SM AI platform.
• SimBoard-first vertical slice, using existing SimBoard metadata and existing simulation detail workflows.
• Start with the smallest useful end-to-end path before broader AI platform scope.

Initial capabilities:

• Retrieve structured simulation metadata for a selected simulation.
• Generate a structured summary from available metadata.
• Return key points, caveats, limitations, and suggested follow-up questions.
• Include citations or source references for claims grounded in available metadata and any validated future retrieval context.
• Clearly indicate when information is missing, unavailable, or unsupported.
• Expose the summary through a basic SimBoard UI action and result panel on the simulation detail page.

Initial UI expectation:

• The first UI should not be an open-ended chat window.
• The first UI should be a read-only summary experience tied to the current simulation context.
• Recommended v1 shape: a "Generate AI Summary" action plus a contextual summary panel, card, drawer, or tab section on the simulation detail page.
• The UI should show loading, error, answer, caveats, citations, limitations, and suggested follow-up questions.
• Broad conversational chat should remain later-scope until the summary workflow, source grounding, and safety constraints are validated.

Initial constraints:

• Read-only only: no write actions, job submission, or configuration changes.
• Backend must validate user and simulation access before retrieving or returning content.
• Permission filtering must happen before content is provided to any model.
• LLM credentials, if used, must be managed server-side and must not be exposed in the browser.
• Default reliance on individual user-supplied API keys is out of scope.

Phased delivery:

• Phase 1A: Deterministic metadata summary endpoint using existing SimBoard metadata, without requiring LLM generation.
• Phase 1B: Minimal frontend integration on the simulation detail page so users can generate and review the summary in place.
• Phase 2: Backend-managed LLM summary generation, if an approved access model is available.
• Phase 3: Add a small curated documentation retrieval path for source-grounded context and richer citations.
• Later / optional: Diagnostics explanation, two-run comparison, and broader conversational workflows once the summary workflow is stable and the metadata path is validated.

Current backend/API shape:

• POST /simulations/{sim_id}/summary -> returns structured simulation summary.
• POST /compare -> optional later-scope endpoint for comparative summaries between simulations.

Implementation direction:

• Existing SimBoard FastAPI backend and Pydantic models.
• Existing SimBoard simulation detail page and frontend API/data-fetching patterns.
• Plain Python service class for the first assistant workflow.
• Structured response contract including answer, citations, assumptions, caveats, limitations, suggested follow-ups, and trace ID.
• Basic structured logging for request IDs, source IDs, latency, failures, and feedback.
• Modular provider/prompt design so model choices can evolve without rewriting the feature.

Additional context

This feature supports SimBoard’s goal of surfacing key insights from E3SM and related Earth system simulations while validating patterns that may later inform broader E3SM AI work.

Integration targets:

• Existing SimBoard FastAPI backend.
• Existing SimBoard metadata models.
• Existing simulation detail workflows in the frontend.
• Future database persistence or caching only after the baseline prototype proves useful.
• Future retrieval, citation, comparison, and broader chat-style interactions only after the baseline summary workflow is validated.

Related Product Requirements Document