app.py

import json
import os
import warnings
from typing import List
from operator import itemgetter
from dotenv import load_dotenv

from langfuse.callback import CallbackHandler
langfuse_handler = CallbackHandler(
    public_key="pk-lf-b5beeffa-b30f-4390-b850-5a1628c7bc5e",
    secret_key="sk-lf-a3e49606-e293-4d32-b451-21627a02556e",
    host="https://cloud.langfuse.com"
)

import gradio as gr

from langchain_openai import OpenAIEmbeddings
from langchain_openai import ChatOpenAI

from langchain.docstore.document import Document

from langchain.prompts import PromptTemplate
from langchain.prompts import ChatPromptTemplate

from langchain.chains import RetrievalQA

from langchain_core.output_parsers import StrOutputParser, CommaSeparatedListOutputParser
from pydantic import BaseModel, Field
from langchain_core.runnables import RunnablePassthrough
from langchain_core.runnables import RunnableParallel

from langchain_community.vectorstores import FAISS
from langchain_community.document_transformers import EmbeddingsRedundantFilter

from langchain.retrievers import EnsembleRetriever
from langchain.retrievers.merger_retriever import MergerRetriever
from langchain.retrievers.document_compressors import DocumentCompressorPipeline
from langchain.retrievers import ContextualCompressionRetriever

#from langchain_cohere import  CohereEmbeddings, CohereRerank

#from langchain_groq import ChatGroq

load_dotenv()

#os.getenv("COHERE_API_KEY")
#os.getenv("GROQ_API_KEY")
os.getenv("OPENAI_API_KEY")

embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
#embeddings_cohere = CohereEmbeddings(model="embed-multilingual-v3.0")

def load_vectorstore(index_name, embeddings, k=20):
    return FAISS.load_local(index_name, embeddings, allow_dangerous_deserialization=True).as_retriever(
                            search_kwargs={ "k": k}
    )

retriever_names = ['large']

retrievers = {}
retrievers_docs = {}

for name in retriever_names:
    retrievers[name] = load_vectorstore(f"{name}", embeddings)
    retrievers_docs[name] = (lambda x: x["input"]) | retrievers[name]


def format_docs_with_id(docs: List[Document]) -> str:
    """
    """
    formatted = [
        (
            f"Metadata: {doc.metadata}\n"
            f"Content: {doc.page_content}\n"

        )
        for doc in docs
    ]
    return "\n\n" + "\n\n".join(formatted)


def prompt_fn():

    return (
            "You are an expert pharmachemist, answer the question based on the context. Do not answer anything not related to pharmachemistry."
            "\n\nHere is the context: "
            "{context}"
            )

llm = ChatOpenAI(temperature=0, model="gpt-4o")

retrieve_docs = {name: (lambda x: x["input"]) | retrievers[name] for name in retriever_names}

def legal(question):

    prompt = ChatPromptTemplate.from_messages([
    ("system", prompt_fn()),
    ("human", "{input}"),
    ])


    rag_chain_from_docs = (
        RunnablePassthrough.assign(context=(lambda x: format_docs_with_id(x["context"])))
        | prompt
        | llm
    )
          
    chains = {
        name: RunnablePassthrough.assign(context=retrieve_docs[name]).assign(answer=rag_chain_from_docs)
        for name in retriever_names
    }

    name = "large"
    
    if name not in chains:
        raise ValueError(f"Invalid typologie: {name}")
    #try:
    #result = chains[name].invoke({"input": question})
    result = chains[name].invoke({"input": question}, config={"callbacks": [langfuse_handler]})
    return result["answer"].content #result["answer"].articles, result["answer"].citations
    #except Exception as e:
    #    return "Je ne sais pa#| "", ""
        

with gr.Blocks() as demo:
    #gr.Markdown("## OnScent Fragrance Intelligent Library Search")
    #gr.Markdown("Developed by ScentGenie")
    #gr.Markdown("### Client Brief or Description")
    
    with gr.Row():
        input1 = gr.Textbox(label="Question", placeholder="effets indesirables du paracetamol")
    #gr.Markdown("### Additional Criterias")
    #gr.Markdown("Criterias like Application area, RMC, Notes to exclude etc")
    #with gr.Row():
    #    input2 = gr.Textbox(label="Additional Criterias (can be left empy)", placeholder=" for hair products with RMC under 15$ and without vanilla note")
    #with gr.Row():
    #    input3 = gr.Dropdown(["Advanced", "Fast"], label="Mode", value="Advanced")
        
    #gr.Markdown("## Recommended Formulas")
    output1 = gr.Text(label="Reponse")
    #output2 = gr.Text(label="Documents Sources")
    #output3 = gr.Text(label="Documents IDs")

    btn = gr.Button("Submit")
    btn.click(legal, inputs=[input1], outputs=[output1])

demo.launch(debug=True)