Roadmap and Milestones

Roadmap & Milestones

Blinkql is built in milestones. Each one ends with a working, deployed app and, where relevant, published benchmark results. Progress is tracked in GitHub Milestones and issues.

Legend: ✅ Done · 🛠️ In progress · 🔜 Planned

flowchart LR
  M0["M0<br/>Foundations"] --> M1["M1<br/>Core query loop"] --> M2["M2<br/>Perf pass I"] --> M3["M3<br/>Explore & visualize"]
  M3 --> M4["M4<br/>Big files & persistence"] --> M5["M5<br/>Operations pack"] --> M6["M6<br/>Perf pass II"] --> M7["M7<br/>v1.0"]

Loading

M0: Foundations ✅

Goal: a solid base so every later change is fast, safe and measurable.

• Vite + React + TypeScript (strict) scaffold with shadcn/ui and Tailwind CSS
• Biome, Vitest and Playwright configured
• GitHub Actions CI: typecheck, lint, test, build
• Vercel production and preview deployments, with cross-origin isolation headers
• Benchmark harness skeleton and deterministic dataset generator
• Issue and PR templates, CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md
• Landing page that says what Blinkql is, says plainly that it's a work in progress, and invites contributors, designed to the Design Language (#72)

Exit criteria: CI green on main; app shell live at blinkql.vercel.app; landing page live, showing work-in-progress status and inviting contributors; pnpm bench runs end to end on a stub scenario.

M1: Core query loop (intentionally naive) ✅

Goal: a working workbench, and an honest baseline.

• Load DuckDB-wasm; import CSV, Parquet and JSON by drag and drop
• SQL editor with a run shortcut, a results table and a table list
• Readable error messages
• A deliberately simple implementation: results converted to row objects, kept in React state and rendered as a plain table

Exit criteria: a 1M-row file queried end to end; baseline benchmark (P-01) published.

M2: Performance pass I, never block the UI ✅

• Results kept as Arrow, with lazy cell reads (no row-object conversion)
• Results stored outside React state
• Virtualized results grid
• Streaming record batches
• Query cancellation and superseded-query handling

Exit criteria: no main-thread task over 50 ms during queries; write-ups P-02 to P-06 published.

M3: Explore & visualize ✅

• Schema browser with column types and statistics; one-click profiling (SUMMARIZE)
• Schema-aware autocomplete
• Chart builder (bar, line, scatter, histogram, Pareto) with SQL-side aggregation
• Query history and saved queries
• Export to CSV, Parquet and PNG

Exit criteria: in a usability check with 3–5 people, someone who doesn't write SQL can profile a file and build a chart in under a minute.

M4: Big files & persistence 🛠️

• OPFS persistence: reopen previously imported datasets instantly
• Optional CSV → Parquet conversion on import
• Remote Parquet and CSV by URL
• Memory guardrails: size warnings, graceful out-of-memory recovery, engine restart

Exit criteria: previous datasets available immediately after a reload; a remote Parquet schema inspected without a full download.

M5: Operations analytics pack 🔜

• Synthetic manufacturing sample datasets: telemetry, breakdowns, shifts, quality
• Templates: OEE, downtime Pareto, MTTR/MTBF, shift-aware reporting (including night shifts across midnight), two-file reconciliation, data-quality checks for import files
• Column mapping, so templates run against any user's file

Exit criteria: every template runs on the sample data and on a user file with mapped columns.

M6: Performance pass II, cold start & rendering 🔜

• Lazy engine loading after the app shell; self-hosted engine with long-term caching
• Service worker caching and offline mode (PWA)
• Canvas-rendered results grid
• Single-threaded vs multithreaded engine evaluation
• Code-split editor and charts
• Performance regression checks in CI

Exit criteria: all performance budgets met; CI fails on budget regression.

M7: v1.0 🔜

• Accessibility pass: keyboard-only flows, screen-reader labels, axe-clean core flows
• Keyboard shortcuts and a command palette
• Shareable query links (SQL and chart settings only, never data)
• Onboarding tour using the sample datasets
• Complete documentation and v1.0 release notes

Exit criteria: the success metrics are met; v1.0 tagged.

Beyond 1.0: ideas under discussion

Not committed. Open for discussion and contributions:

• Natural-language-to-SQL assistant, strictly opt-in, using a bring-your-own API key or an in-browser model, with a clear disclosure of exactly what is sent
• Notebook-style analyses with multiple query and chart cells
• Export an analysis as a self-contained HTML report
• Excel (.xlsx) import
• Geospatial queries with the DuckDB spatial extension, for plant layouts and logistics
• The results grid as a standalone, embeddable package
• More template packs: logistics, retail, finance, workforce analytics

Have an idea? Start a thread in Discussions.