=== requirements.txt
pandion[lancedb,local,pdfium]~=<pin>

# Uncomment the next line to install the much smaller CPU-only build of torch.
# --extra-index-url https://download.pytorch.org/whl/cpu
=== README.md (head)
# project

This folder runs a Pandion retrieval pipeline on your own documents. The pipeline is in `pipeline.yaml`.

Every step in this pipeline runs on your computer, so it needs no API key.

=== pipeline.yaml
path: ./pandion-index
index:
  parse:
    use: pdfium
    join_lines: true
  clean:
  - use: header_footer_strip
    min_page_ratio: 0.5
  - use: dedupe_blocks
  chunk:
    use: sentence_window
    sentences_per_chunk: 5
    overlap_sentences: 1
  embed:
    use: qwen3-embedding-0.6b
  store:
    use: lancedb
    metric: cosine
    build_fts: true
retrieve:
  search:
    use: hybrid_rrf
    top_k: 30
    rrf_k: 60
    query_expansion: none
  rerank:
  - use: cross_encoder
    model: BAAI/bge-reranker-v2-m3
    top_k: 5
answer:
  use: search
  top_k: 5
