import pandas as pd
from sentence_transformers import SentenceTransformer, util

model = SentenceTransformer("all-MiniLM-L6-v2")

def get_relevant_passages(query, df, top_k=20):
    corpus = df["description"].astype(str).tolist()
    corpus_embeddings = model.encode(corpus, convert_to_tensor=True)
    query_embedding = model.encode(query, convert_to_tensor=True)

    hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=top_k)[0]
    return df.iloc[[hit['corpus_id'] for hit in hits]]