tools.py

import json
from dotenv import load_dotenv
from langchain_core.documents import Document
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain.tools import BaseTool, StructuredTool, tool
from langchain.chains.query_constructor.schema import AttributeInfo

load_dotenv()

# Create knowledge of app functions
docs = [
    Document(
        page_content="""
        An app that aims to enhance spatial navigation for visually impaired individuals by utilizing an intelligent AI-driven system that aids in grocery shopping.
        At the core of this is a bracelet equipped with vibration feedback. By varying the intensity of the vibration, the app guides users to products in a supermarket.
        The app is designed to interface with the bracelet. The goal is to empower blind users with greater independence and ease in their supermarket shopping experiences.
        """,
        metadata={"app_id": 1, "name": "Spatial Navigator"},
    ),
    Document(
        page_content="""
        A list where a user can add and remove products from that they want to aquire at a supermarket.
        The app will send you the current grocery list as additional data.
        Return the product(s) that need to be added, appended with either + (to add) or - (to remove), including the amount.
        Example: [+4]Jonagold Apple; [-1]Quaker Oats""",
        metadata={"app_id": 1, "type": "function", "label": "grocery-list", "additional-data-required": True},
    ),
    Document(
        page_content="Enables the user to scan barcodes of products to gain detailed nutritional information about them.",
        metadata={"app_id": 1, "label": "barcode-scanner", "additional-data-required": False},
    ),
    Document(
        page_content="""
        Allows the user to save their favorite supermarkets and start GPS navigation to this supermarket.
        Extra Feature: If the user mentions a name of a supermarket, return the name of the supermarket.""",
        metadata={"app_id": 1, "label": "navigation", "additional-data-required": False},
    ),
    Document(
        page_content="Can identify products in a supermarket and hands of the user in real time to help find the right products.",
        metadata={"app_id": 1, "label": "object-and-hand-recognition", "additional-data-required": False},
    ),
]

additional_data_descriptions = {
    "app_id": 1,
    "label": "grocery-list",
    "instruction": """
        Extra Feature: Return the product(s) that need to be added, appended with either + (to add) or - (to remove), including the amount.
        Example: [+4]Jonagold Apple; [-1]Quaker Oats"""
}

# metadata_field_info = [
#     AttributeInfo(
#         name="type",
#         description="Whether this document describes the app itself or a functionality within the app.",
#         type="string",
#     ),
#     AttributeInfo(
#         name="name",
#         description="The name of the app.",
#         type="string",
#     ),
#     AttributeInfo(
#         name="app_id",
#         description="The unique identifyer of the app. Connects app descriptions to app functionality.",
#         type="integer",
#     ),
#     AttributeInfo(
#         name="label",
#         description="The unique label of a function within an app.",
#         type="string",
#     )
# ]

embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = Chroma.from_documents(docs, embeddings)

app_descriptions_retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 3, "filter": {"type": "app"}},
)

app_functions_retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 3, "filter": {"type": "function"}},
)

@tool
def app_functions(query: str) -> str:
    """Gives the most relevant app functions including descriptions of it's purpose and accompanying metadata based on the input string."""
    docsList = app_functions_retriever.invoke(query)
    strList =[]
    # Export the document content as well as its metadata
    for doc in docsList:
        strList.append({"document" : doc.page_content ,"metadata" : doc.metadata})
    return json.dumps(strList)