Feat: Visualizations, Dashboards, Model Registry CLI, Project Filtering, Search, Notifications & E2E Tests

[Jovonni]

addresses #3 #4 #5

Major feature release adding 6 new subsystems and comprehensive fixes across the full stack. ~11,400 lines added across 83 files spanning the Rust API, Python SDK, Next.js frontend, model-runner, Kubernetes deployment, and E2E test suite.

• Visualizations & Dashboards — Full visualization system with 9 backends (matplotlib, plotly, altair, seaborn, bokeh, datashader, networkx, geopandas, plotnine), unified render()
  abstraction, in-browser Monaco editor with live preview, and drag-and-drop dashboard builder with panel layout persistence
• Model Registry CLI ↔ UI integration — Bidirectional sync between CLI openmodelstudio install/uninstall and UI badges. use_model() SDK function resolves models via API (works
  inside K8s workspace pods). Auto-detection of local platform API + auto-login with default credentials
• Project-scoped filtering — Every noun (models, datasets, experiments, training jobs, workspaces, features, monitoring, visualizations, data sources, AutoML) now filters by
  selected project via ProjectFilterProvider in the topbar
• Global search — Cmd+K overlay searching across models, datasets, experiments, training jobs, projects, and visualizations with instant navigation
• Notifications system — Real-time notification bell with unread count polling, grouped timeline (Today/This Week/Earlier), mark-all-read, and context-aware icons per notification
  type
• Training job duration fix — Fixed duration showing "0m" or "0:00" after completion by preserving original started_at timestamp, adding seconds to duration display, marking jobs
  completed when progress hits 100%, and adding live polling on list/detail pages
• Workspace tutorial notebooks — Three pre-loaded notebooks (welcome.ipynb, visualization.ipynb, registry.ipynb) covering the full ML workflow, all 9 viz backends, and registry SDK
  usage

Changes by area

Backend (Rust API)

• New visualizations routes — full CRUD + publish endpoint for visualizations
• New notifications routes — unread count, mark read, mark all read
• New search route — multi-entity search across the platform
• New resolve-registry endpoint — resolve models by registry_name for SDK use_model()
• New registry-uninstall endpoint — unregister models from platform on CLI uninstall
• New registry-status endpoint — batch check install status for UI badges
• Project filter (?project_id=) added to all list endpoints (models, datasets, experiments, training, workspaces, features, monitoring, data sources, AutoML)
• Fixed nullable project_id across Model, Dataset, DataSource, Pipeline structs (Uuid → Option) — was causing 500 errors on any SELECT of models without a project
• post_metrics now sets status = 'completed' and completed_at when progress reaches 100%
• DB migrations: 003_visualizations.sql (visualizations + dashboards tables), 004_registry_name.sql

Python SDK

• New visualization.py — create_visualization(), render() (auto-detects backend), publish_visualization(), dashboard CRUD
• New registry.py — registry_search(), registry_list(), registry_info(), registry_install(), registry_uninstall(), list_installed(), set_registry() with auto-detect API + auto-login
• New client.py — RegistryModel class, use_model() method, register_model() enhanced to accept RegistryModel instances
• New cli.py — Full CLI (openmodelstudio search/registry/info/install/uninstall/list/config)
• New config.py — Persistent config (~/.openmodelstudio/config.json), registry URL + models dir management
• init.py — Exports all new SDK functions

Frontend (Next.js)

• New pages: /visualizations, /visualizations/[id], /dashboards, /dashboards/[id], /registry, /registry/[id], /search
• viz-renderer.tsx — Renders Plotly (via Plotly.js), Altair (via vega-embed), Bokeh (via BokehJS), SVG, and PNG outputs in-browser
• notification-panel.tsx — Popover with grouped notifications, unread badge, scroll area, fully opaque background
• search-overlay.tsx — Cmd+K command palette with multi-entity search
• project-filter.tsx + project-filter-provider.tsx — Global project scope selector in topbar
• All list pages updated to use api.getFiltered() with project scope
• Training detail page: 3-second polling for live metrics, loss/accuracy charts auto-update
• Training list page: duration now shows seconds (45s not 0m), live polling + per-second re-render for running jobs
• Registry detail page: install/uninstall button, Installed/Not Installed badges
• Sidebar restructured with Develop/Operate/Admin sections
• Fixed infinite reload loop in ProjectFilterProvider
• Fixed shouldFilter={false} on cmdk CommandInput for proper search behavior

Model Runner

• runner.py: Changed started_at = NOW() → started_at = COALESCE(started_at, NOW()) to preserve the original timestamp set when the K8s job was created, preventing duration from
  collapsing to near-zero on fast-training models

Kubernetes / Deployment

• Dockerfile.workspace updated to include all 3 tutorial notebooks + all 9 visualization backend packages (plotly, altair, bokeh, plotnine, networkx, geopandas, datashader)

Workspace Notebooks

• welcome.ipynb — Extended from 13 to 16 cells (added visualization creation, interactive Plotly chart, dashboard building)
• visualization.ipynb — New 25-cell notebook covering all 9 visualization backends, render() reference, dynamic visualizations, and dashboard SDK
• registry.ipynb — New 21-cell notebook covering registry search, install, use_model(), register + train, uninstall, and CLI reference

Documentation

• docs/VISUALIZATIONS.md — Complete visualization & dashboards reference (9 backends, render() function, in-browser editor, dashboard SDK, API reference)
• docs/MODELING.md — Extended with cells 14-16 (visualization, Plotly, dashboards)
• docs/CLI-REGISTRY.md — Full CLI & SDK reference for model registry

Tests

• search-and-registry.spec.ts — 14 Playwright tests for search overlay and registry page
• dashboard-verification.spec.ts — Dashboard page E2E tests
• notebook-workflow.spec.ts — Workspace notebook workflow tests
• notifications-search.spec.ts — Notifications and search integration tests
• project-filter-flow.spec.ts — Project filtering E2E tests
• tests/python/sdk/test_cli.py — CLI unit tests
• All 14 E2E tests passing

Testing Strategy

• cargo check — Rust API compiles clean
• npx tsc --noEmit — Frontend compiles clean
• All 14 Playwright E2E tests pass
• CLI install → UI shows "Installed" badge on registry page
• CLI uninstall → UI reverts to "Not Installed" badge
• oms.use_model("iris-svm") works inside workspace containers
• oms.register_model("my-iris", model=iris) succeeds
• Training job duration displays correctly after completion (seconds-level accuracy)
• Training job metrics auto-update without page refresh
• Notifications dropdown is scrollable, opaque, self-contained
• Global search (Cmd+K) finds models, datasets, experiments, jobs
• Project filter scopes all pages correctly
• Visualization render() works for matplotlib, plotly, altair backends
• Dashboard panels persist layout on save