menu_chromadb_semantic_search.py

import chromadb
from chromadb.utils import embedding_functions
import csv

# --- Setup ChromaDB (in-memory for Hugging Face Spaces free tier) ---
chroma_client = chromadb.Client()

# SentenceTransformer embedding function
sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(
    model_name="all-mpnet-base-v2"
)

# Create collection
collection = chroma_client.create_collection(
    name="my_collection",
    embedding_function=sentence_transformer_ef
)

# --- Load CSV data ---
with open("menu_items.csv") as file:
    lines = csv.reader(file)
    documents = []
    metadatas = []
    ids = []

    for i, line in enumerate(lines):
        if i == 0:
            continue  # skip header
        documents.append(line[1])
        metadatas.append({"item_id": line[0]})
        ids.append(str(i))

# Add to ChromaDB
collection.add(
    documents=documents,
    metadatas=metadatas,
    ids=ids
)

def search_dishes(query: str) -> str:
    """Search for top 5 similar dishes."""
    results = collection.query(
        query_texts=[query],
        n_results=5,
        include=["documents", "metadatas"]
    )
    hits = results["documents"][0]
    ids_meta = results["metadatas"][0]
    output = [f"{ids_meta[i]['item_id']}: {hits[i]}" for i in range(len(hits))]
    return "\n".join(output)