v3.1.0 — LangChain integration, BEIR benchmarks, biomedical preset

What's new in v3.1.0

LangChain integration

VORTEXRAG now works as a drop-in BaseRetriever in any LangChain pipeline:

from integrations.langchain_retriever import VortexRAGRetriever

retriever = VortexRAGRetriever(domain="medical", top_k=5)
retriever.add_documents(your_docs)

# works with RetrievalQA, ConversationalRetrievalChain, LCEL, etc.
docs = retriever.invoke("What causes sepsis?")

BEIR benchmark evaluation script

Run VORTEXRAG against the full BEIR benchmark suite:

pip install beir
python benchmarks/eval_beir.py --datasets nq hotpotqa scifact

Biomedical domain preset

New biomedical domain (τ=0.32) tuned for PubMed/BioASQ literature retrieval — sits between scientific (0.30) and medical (0.35) in strictness.

Branch structure

• main — protected, stable
• dev — integration branch, all contributor PRs target this
• feat/* — feature branches

247 tests passing across Python 3.10–3.13

---

Paper: https://doi.org/10.5281/zenodo.20579702