retriever.py

from embedder import get_model

def retrieve_chunks(index, texts, question, top_k=15):
    model = get_model()
    q_embedding = model.encode([question], convert_to_numpy=True, normalize_embeddings=True)[0]
    # Use Pinecone v3 index query
    res = index.query(vector=q_embedding.tolist(), top_k=top_k, include_metadata=True)
    selected_texts = [match['metadata']['text'] for match in res['matches']]
    return selected_texts