ai-sandbox/ident/ident_v2.py

from langchain.document_loaders.csv_loader import CSVLoader
from langchain_community.vectorstores import Chroma
from langchain_community.embeddings import OllamaEmbeddings
from langchain_community.chat_models import ChatOllama
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain
from langchain.text_splitter import RecursiveCharacterTextSplitter


class IdentV2:
    loader = None
    documents = None
    embeddings = None
    vectorstore = None
    llm = None

    def __init__(self):
        self.loader = CSVLoader(file_path="../csv/bm_res_lite_peugeot.csv", metadata_columns=["modelo_id"])
        self.documents = self.loader.load()
        print(f"Loaded {len(self.documents)} registers")
        self.embeddings = OllamaEmbeddings(model="nomic-embed-text")

        chunks = RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=0).split_documents(self.documents)
        self.vectorstore = Chroma.from_documents(chunks, self.embeddings, persist_directory="db/chroma_db")
        self.llm = ChatOllama(model_name="mistral",  temperature=0)

    def search(self, query, k=7):
        docs = self.vectorstore.similarity_search(query, k=k)

        print(f"Search results: {docs}")

        return [(doc.page_content, 'modelo_id: ' + doc.metadata["modelo_id"]) for doc in docs]

    def get_chain(self):
        template = """
        You are a world expert in vehicle appraisal
        Below I am going to share with you a partial description of a vehicle model:
        {query}
        
        Your job is to try to find the vehicle model that best fits the description from the following list of available models:
        {models}
        
        Consider the following comparison criteria in descending order:
        - Year
        - Brand
        - Model
        - Transmission
        - Variant
        - Displacement
        - Fuel
        - Traction
        - Number of doors
        - Price (10%)
        
        Explain your comparison criterion and give me a list of the models ordered in descending order that have best fitted, including de 'model_id'.
        """
        prompt = PromptTemplate(input_variables=["models", "query"], template=template)
        chain = LLMChain(llm=self.llm, prompt=prompt)

        return chain

    def ident(self, query):
        results = self.search(query)
        chain = self.get_chain()
        response = chain.invoke(dict(query=query, models=results))
        return response['text']


print(IdentV2().ident("PEUGEOT 208 2020 1.2 SIGNATURE 1.2 PURETECH 82 MT KILOMETRAJE 65035 KM TRANSMISION MT PUERTAS 5 ANO 2020 TRACCION 4X2 MOTOR 1200 LT TIPO HATCHBACK COMBUSTIBLE BENCINA COLOR GRIS PLATINUM"))