alirezaaminzadeh/or-knowledge-copilot-retriever

OR Knowledge Copilot Retriever

Hybrid BM25 + word TF-IDF + character TF-IDF index for multi-layer operations-research knowledge (formulation, code, solver traces).

This artifact is a fitted lexical index, not a fine-tuned transformer. It is intended for CPU Spaces and on-prem retrieval without GPU.

Files

File	Role
hybrid_index.joblib	Vectorizers, TF-IDF matrices, BM25 statistics
config.json	Vocabulary sizes and chunk count

Usage

from ragkit.pipeline import RAGPipeline
from domain.corpus import SYNONYMS, router, PRODUCT

pipe = RAGPipeline(
    chunks_path="dataset/chunks.jsonl",
    model_path="model/hybrid_index.joblib",
    synonyms=SYNONYMS,
    product=PRODUCT,
    router=router,
)
print(pipe.ask("What spinning reserve does GridWest unit commitment enforce?").answer)

Evaluation

See eval_results.json on the companion dataset alirezaaminzadeh/or-knowledge-copilot-corpus.