Show HN: Selecta – Natural language analytics for BigQuery (github.com)

Selecta is an open-source demo that provides natural-language analytics for BigQuery by combining Google’s ADK (running Gemini or Vertex AI) with a streaming Next.js frontend. Users ask questions in plain English, the ADK agent generates and runs GoogleSQL, and results stream back as a structured payload (intermediate reasoning + final sections). Every response includes Summary, Results, and Business Insights, a Vega-Lite chart spec, rows/columns/rowCount for table previews, and BigQuery metadata (executionMs, jobId, dataset), all surfaced in a responsive Next.js 14 UI with tabs for Chart / Table / SQL / Meta and a “View reasoning” accordion. Technically, Selecta uses an SSE-based POST /run_sse flow: Next.js (Node 22+, TypeScript, Zustand, Shadcn UI, Tailwind 4) acts as the SSE client, the ADK HTTP server (Python 3.12+) streams events according to backend/api-contract.md, and BigQuery executes the queries. The repo ships with a sample dataset, a Cloud Run deploy script, and a drop-in ADK backend (no custom FastAPI layer). For ML practitioners this is significant: it gives end-to-end LLM→SQL reproducibility, real-time reasoning traceability, execution telemetry for auditing and debugging, and a clear event schema you can extend—making it easier to build observable, production-ready NL→analytics workflows. Apache 2.0 licensed; docs and roadmap included.