Skip to content

v0.3.0

@TheYonk TheYonk tagged this 30 Apr 13:51
First public release. Single-cell pipeline + bakeoff at parity across
Python and Rust. Same canonical YAML runs from either implementation
and produces equivalent leaderboards.

Highlights:
- Rust port at parity (single-cell + bakeoff). Orchestrator stays
  Python-only.
- YAML-driven HuggingFace embedder pointer ("BYO embedder") in both
  languages, with mean-pooling support.
- Bakeoff leaderboard surfaces speed-vs-quality (chunks / ingest_s /
  embed_s + per-embedder query-time cost).
- Inline (library) mode: chunkshop.Pipeline (Python) and
  chunkshop::Pipeline (Rust).
- pg_table source: metadata_columns + VIEW pattern for JOINed
  metadata.
- target.delete_orphans for atomic per-doc shrink cleanup.
- Bundled sample corpora (NTSB + sales-crm), compressed in-tree.
- User-journey-first docs.

Full ledger in CHANGELOG.md.
Assets 2
Loading