Build spec

Chaveta is being built as a generator platform, not a dataset warehouse.

The product plan and concrete build spec both center the generator package, validation trace, repair logic, exports, and quality reporting as the durable asset.

Primary routes
/, /research, /product, /contact, then /app and run detail pages for operators.
Core packages
UI, config, db, schemas, ai, domain, workflows, generators, exporters, validation, observability.
Design posture
Beag-inspired editorial framing, monospaced display tone, high readability, dark and light mode from day one.
v1 boundaries

Will do

  • Accept natural-language generation requests.
  • Classify into one primary training type.
  • Compile a machine-readable generation spec.
  • Author a generator package and run smoke-batch generation.
  • Validate, repair, export, and display a quality report.

Will not do

  • Support arbitrary user code execution.
  • Offer unlimited export formats.
  • Blend multiple training families into one run by default.
  • Pretend benchmark uplift is guaranteed for every domain.