app.py

# NonToxicGlazeAdvisor_Chat_with_Docs_Groq_Edition_1 - app.py - 04-04-2024

# STREAMLIT:
# https://www.datacamp.com/tutorial/streamlit:
#
# st.title(): This function allows you to add the title of the app. 
# st.header(): This function is used to set header of a section. 
# st.markdown(): This function is used to set a markdown of a section. 
# st.subheader(): This function is used to set sub-header of a section. 
# st.caption(): This function is used to write caption. 
# st.code(): This function is used to set a code. 
# st.latex(): This function is used to display mathematical expressions formatted as LaTeX.
# 
# st.title ("this is the app title")
# st.header("this is the header ")
# st.markdown("this is the markdown")
# st.subheader("this is the subheader")
# st.caption("this is the caption")
# st.code("x=2021")
# st.latex(r''' a+a r^1+a r^2+a r^3 ''')


# ========================================================================
# TABS
# 03-04-2024
import streamlit as st

tab1, tab2, tab3 = st.tabs(["AI Adviseur voor niet-giftige glazuren", "Uitleg", "Voorbeeld vragen"])

#with tab1:
#   st.header("A cat")
#   st.image("https://static.streamlit.io/examples/cat.jpg", width=200)

with tab2:
    st.header("Uitleg")
    st.image("De fritte oven - een schets van het proces.jpg", width=200)

with tab3:
    st.header("Voorbeeld vragen")
    st.write("De fritte oven - een schets van het proces:")
    st.image("De fritte oven - een schets van het proces.jpg", width=200)

    # https://docs.streamlit.io/library/api-reference/widgets/st.text_area
    # st.text_area
    # Streamlit Version -Version 1.32.0
    # Display a multi-line text input widget.


    # txt = st.text_area() # is voor text input !
    # st.header('This is a header with a divider', divider='rainbow')
    st.header('Hier alvast een klein aantal voorbeeldvragen die je aan deze AI adviseur voor niet-giftige glazuren zou kunnen stellen:', divider='rainbow')
    # st.write("**Hier alvast een klein aantal voorbeeldvragen die je aan deze AI adviseur voor niet-giftige glazuren zou kunnen stellen:**")
    st.write("\n")
    st.write("Hoe kan ik thuis als hobby keramiek maken? Geef het antwoord uitsluitend en alleen in de Nederlands taal  ")
    st.write("Geef een samenvatting in 10 zinnen. ")
    st.write("Geef een samenvatting in 25 bullet points. ")
    st.write("Geef een samenvatting in 25 bullet points. Geef het antwoord uitsluitend en alleen in de Nederlandse taal. ")
    st.write("Geef een samenvatting in 25 bullet points. Geef het antwoord uitsluitend en alleen in de Duitse taal. ")
    st.write("Explain glazing techniques to me as if I am 10 years old. ")
    st.write("Geef me uitleg over glazuurtechnieken alsof ik 5 jaar oud ben. Geef het antwoord uitsluitend en alleen in de Nederlandse taal.")

    st.write(" ... ")

    st.header('Voorbeeldvragen met voorbeeldantwoorden die je aan deze AI adviseur voor niet-giftige glazuren zou kunnen stellen:', divider='rainbow')

    st.write("Wie ben ik? (selecteer als taal Nederlands)\n")
    st.write("assistant avatar:")
    st.write("Gezien de tekst is het waarschijnlijk dat u heer Victor Benckhuijsen bent. U bent iemand die ervaring heeft opgedaan met het maken van glazuren voor keramiek, met name alkalische fritten. U heeft deze ervaring opgedaan op verschillende plekken, zoals de kunstacademie, de agrarische hogeschool en de geologie en bodemkunde. U benadrukt de waarden van veiligheid en het belang van het gebruik van non-giftige materialen. U verkoopt geen recepten, maar verstrekt wel veiligheidsattesten op verzoek. U benadrukt ook de voordelen van het gebruik van industriële fritten voor hobbyisten, omdat dit een veilige en eenvoudige manier is om non-giftige keramische glazuren te maken.")

    st.write(" ... ")

    st.header('Geavanceerde voorbeeldvragen met voorbeeldantwoorden die je aan deze AI adviseur voor niet-giftige glazuren zou kunnen stellen:', divider='rainbow')

    st.write("Formulate 10 diverse hypotheses related to the context and then find arguments that support and refute these hypotheses  (selecteer als taal Engels)\n")
    st.write("assistant avatar:")
    st.write("""Formulate 10 diverse hypotheses related to the context and then find arguments that support and refute these hypotheses 


RESPONSE :

Response time: 0.42764801399999897 seconds

Hypothesis 1: The use of the Seger formula for representing the chemical structure of glazes in ceramics is a more accurate and reliable method than previous methods.

Argument for: The Seger formula represents the chemical structure of glazes in a way that is consistent with the periodic table, dividing the components into three categories: melting point depressants, amphoteric metals, and fusible metals. This systematic approach allows ceramicists to better understand the relationships between different components and their effects on the final product.

Argument against: While the Seger formula provides a systematic approach, it may not account for all the complexities and subtleties of ceramic glaze formulations. Other factors, such as particle size, impurities, and firing conditions, can also significantly impact the final product.


Hypothesis 2: The addition of Zirkoonsilicaat to a glaze recipe can result in a wider range of pastel colors.

Argument for: Zirkoonsilicaat has been identified as a component that can create pastel colors when added to a glaze recipe. This is because it can act as an opacifier, modifying the way light interacts with the glaze and leading to softer, pastel shades.

Argument against: The addition of Zirkoonsilicaat may also have unintended consequences, such as altering the viscosity or melting point of the glaze, which could impact its overall appearance and performance.


Hypothesis 3: The use of "gefritte" ceramic color pigments can result in more vibrant and consistent colors.

Argument for: "Gefritte" ceramic color pigments are created through a process of heating and cooling, which can result in more consistent and vibrant colors than other pigment types. This is because the heating and cooling process can help to ensure that the pigment particles are evenly distributed throughout the glaze.

Argument against: The use of "gefritte" pigments may not be suitable for all types of ceramic glazes, as the heating and cooling process can also alter the properties of the glaze itself. Additionally, the cost of "gefritte" pigments may be higher than other pigment types.


Hypothesis 4: The practice of alchemy, as described in the context, was a precursor to modern scientific methods.

Argument for: The alchemists described in the context were engaged in a systematic process of experimentation and observation, using a combination of objective and subjective methods. This approach bears some similarities to modern scientific methods, which emphasize careful observation, experimentation, and the development of hypotheses.

Argument against: However, the alchemists' approach was also fundamentally different from modern scientific methods in that it was heavily influenced by spiritual and mystical beliefs. While the alchemists did engage in systematic observation and experimentation, they also placed a strong emphasis on the role of the individual alchemist's subjective experiences and intuitions.


Hypothesis 5: The division of alchemy into "materiële" and "geestelijke" paths was a result of the increasing specialization and professionalization of scientific inquiry.

Argument for: As scientific inquiry became more specialized and professionalized, it became increasingly difficult for individual researchers to maintain a holistic perspective that integrated both material and spiritual concerns. The division of alchemy into "materiële" and "geestelijke" paths may have been a response to this trend, allowing researchers to specialize in one area or the other.

Argument against: However, it is also possible that the division of alchemy into "materiële" and "geestelijke" paths was a result of broader cultural and philosophical trends, rather than simply a response to the professionalization of scientific inquiry.


Hypothesis 6: The Seger formula can be used to accurately predict the properties of a ceramic glaze based on its component parts.

Argument for: The Seger formula provides a systematic approach to understanding the relationships between different components of a ceramic glaze, which could potentially be used to predict the properties of the final product.

Argument against: However, the properties of ceramic glazes can be influenced by a wide range of factors, including particle size, impurities, and firing conditions. As a result, it may be difficult or impossible to accurately predict the properties of a ceramic glaze based solely on its component parts.


Hypothesis 7: The use of clay as a starting material for ceramic glazes is a result of its unique physical and chemical properties.

Argument for: Clay has a number of unique physical and chemical properties that make it well-suited for use in ceramic glazes. For example, clay is highly plastic, which means it can be easily shaped and molded. Additionally, clay undergoes a number of physical and chemical changes when heated, which can result in the formation of strong, durable ceramic materials.

Argument against: However, clay is not the only possible starting material for ceramic glazes. Other materials, such as silica or alumina, can also be used to create ceramic glazes with a wide range of properties.


Hypothesis 8: The practice of alchemy was primarily focused on the transformation of base metals into gold.

Argument for: The transformation of base metals into gold was a central goal of many alchemists, who believed that this transformation was a symbol of spiritual enlightenment and transformation.

Argument against: However, the practice of alchemy was not limited to the transformation of base metals into gold. Alchemists were also interested in a wide range of other topics, including medicine, pharmacology, and spirituality.


Hypothesis 9: The properties of ceramic glazes can be altered through the addition of various additives.

Argument for: The properties of ceramic glazes can be significantly altered through the addition of various additives, such as opacifiers, colorants, and fluxes. These additives can modify the way that light interacts with the glaze, as well as its melting point, viscosity, and other properties.

Argument against: However, the addition of additives can also introduce new challenges and complexities into the glaze-making process. For example, some additives may interact with other components of the glaze in unpredictable ways, leading to unexpected results.


Hypothesis 10: The practice of alchemy was primarily focused on the manipulation of physical materials.

Argument for: The practice of alchemy did involve the manipulation of physical materials, such as metals, minerals, and chemicals. Alchemists engaged in a wide range of experimental techniques, including heating, cooling, dissolving, and mixing, in order to transform these materials.

Argument against: However, the practice of alchemy was not limited to the manipulation of physical materials. Alchemists were also interested in spiritual and mystical concepts, such as the unity of matter and spirit, the transformation of the self, and the pursuit of enlightenment.""")


    
    st.write(" ... ")

    
    st.write("(wordt vervolgd)")



    
    
# ========================================================================

    

with tab1:
    st.header("AI Adviseur voor niet-giftige glazuren")
    st.image("De fritte oven - een schets van het proces.jpg", width=200)

    
    # JB:
    # LangChainDeprecationWarning: Importing embeddings from langchain is deprecated. 
    # Importing from langchain will no longer be supported as of langchain==0.2.0.
    # Please import from langchain-community instead:
    # `from langchain_community.embeddings import FastEmbedEmbeddings`.
    # To install langchain-community run `pip install -U langchain-community`.
    from langchain_community.embeddings import FastEmbedEmbeddings

    import os
    import streamlit as st
    from langchain_groq import ChatGroq
    from langchain_community.document_loaders import WebBaseLoader
    # JB:
    from langchain_community.document_loaders import PyPDFLoader
    from langchain_community.embeddings import OllamaEmbeddings

    # JB:
    from langchain_community.embeddings import FastEmbedEmbeddings
    from langchain_community.document_loaders import PyPDFDirectoryLoader

    # JB:
    # File Directory
    # This covers how to load all documents in a directory.
    # Under the hood, by default this uses the UnstructuredLoader.
    from langchain_community.document_loaders import DirectoryLoader
    from langchain_community.document_loaders import TextLoader
    import chardet

    from langchain_community.vectorstores import FAISS
    # from langchain.vectorstores import Chroma
    # from langchain_community.vectorstores import Chroma

    from langchain.text_splitter import RecursiveCharacterTextSplitter
    from langchain.chains.combine_documents import create_stuff_documents_chain
    from langchain_core.prompts import ChatPromptTemplate
    from langchain.chains import create_retrieval_chain
    import time
    from dotenv import load_dotenv

    import glob

    load_dotenv()  #

    groq_api_key = os.environ['GROQ_API_KEY']
    # groq_api_key = "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" # os.environ['GROQ_API_KEY']
    # groq_api_key = "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" # JB OK 24-03-2024
    # print("groq_api_key: ", groq_api_key)

    # st.title("Chat with Docs - Groq Edition :) ")
    # # st.title ("this is the app title")
    # st.title("Non-Toxic Glaze Advisor:")
    # st.subheader("A tool for getting advicgroqe on non-toxic ceramic glazes for earthenware temperature ranges.")
    # st.subheader("Victor Benchuijsen  : (Glaze techniques / Ceramics)")
    # st.subheader("Jan Bours           : Artificial Intelligence / Data Science / Natural Language Processing (ALL RIGHTS RESERVED)")
    # st.write("---------------------------------")
    # st.subheader("Chat with Docs - Using AI: 'mixtral-8x7b-32768' Groq Edition (Very Fast!) - VERSION 1 - March 18, 2024")
    # st.write("---------------------------------")

    st.title("Adviseur voor niet-giftige glazuren:")
    st.subheader("Een gereedschap gebaseerd op Kunstmatige Intelligentie (AI) om advies te krijgen over niet-giftige keramische glazuren voor aardewerk temperatuur bereiken.")
    st.write("---------------------------------")
    st.subheader("Victor Benckhuijsen  : (Glazuur technieken / Keramiek)")
    st.subheader("(ALL RIGHTS RESERVED)")
    st.image('Victor_Benckhuijsen_2.png', caption='Victor Benckhuijsen', width=100)
    # st.subheader("---------------------------------")
    # st.write("---------------------------------")
    st.subheader("Jan Bours           : Artificial Intelligence / Data Science / Natural Language Processing")
    st.subheader("(ALL RIGHTS RESERVED)")
    st.image('Jan_Bours_2.png', caption='Jan Bours', width=100)
    st.write("---------------------------------")
    st.subheader("Chat with Docs - Using AI: 'mixtral-8x7b-32768' Groq Edition (Very Fast!) - VERSION 2 - April 3, 2024")
    st.write("---------------------------------")

    # st.header("LIST OF ALL THE LOADED DOCUMENTS: ")
    st.header("LIJST MET ALLE ACTUEEL GELADEN DOCUMENTEN: ")

    st.write("")
    pdf_files  = glob.glob("*.pdf")
    # word_files = glob.glob("*.docx")
    for file in pdf_files:
    # for file in word_files:
        st.subheader(file)

    st.write("---------------------------------")


    start1 = time.process_time()

    if "vector" not in st.session_state:
    
        st.write("Even geduld a.u.b. ........")
        # st.header("Chunking, embedding, storing in FAISS vectorstore (Can take a long time!).")
        # st.subheader("Wait till this hase been done before you can enter your query! .......")
    
        # st.session_state.embeddings = OllamaEmbeddings() # ORIGINAL
        st.session_state.embeddings = FastEmbedEmbeddings() # JB


        # st.session_state.loader = WebBaseLoader("https://paulgraham.com/greatwork.html") # ORIGINAL
        # st.session_state.docs = st.session_state.loader.load()                           # ORIGINAL
        # https://api.python.langchain.com/en/latest/document_loaders/langchain_community.document_loaders.pdf.PyPDFLoader.html
        # https://python.langchain.com/docs/integrations/document_loaders/merge_doc
        # from langchain_community.document_loaders import PyPDFLoader
        # loader_pdf = PyPDFLoader("../MachineLearning-Lecture01.pdf")
        #
        # https://stackoverflow.com/questions/60215731/pypdf-to-read-each-pdf-in-a-folder
        # 
        # https://api.python.langchain.com/en/latest/document_loaders/langchain_community.document_loaders.pdf.PyPDFDirectoryLoader.html
        # https://python.langchain.com/docs/modules/data_connection/document_loaders/pdf#pypdf-directory 
        # !!!!!
        # PyPDF Directory
        # Load PDFs from directory
        # from langchain_community.document_loaders import PyPDFDirectoryLoader
        # loader = PyPDFDirectoryLoader("example_data/")
        # docs = loader.load()
        #
        # ZIE OOK:
        # https://python.langchain.com/docs/modules/data_connection/document_loaders/pdf#using-pypdf
        # Using MathPix
        # Inspired by Daniel Gross's https://gist.github.com/danielgross/3ab4104e14faccc12b49200843adab21
        # from langchain_community.document_loaders import MathpixPDFLoader
        # loader = MathpixPDFLoader("example_data/layout-parser-paper.pdf")
        # data = loader.load()
        # pdf_file_path = "*.pdf"                                                 # JB
        # st.session_state.loader = PyPDFLoader(file_path=pdf_file_path).load()   # JB
        # st.session_state.loader = PyPDFLoader(*.pdf).load()                     # JB syntax error *.pdf !
        # st.session_state.loader = PyPDFDirectoryLoader("*.pdf")                 # JB PyPDFDirectoryLoader("example_data/")   
        # chunks = self.text_splitter.split_documents(docs)
        # chunks = filter_complex_metadata(chunks)

        # JB:
        # https://python.langchain.com/docs/modules/data_connection/document_loaders/pdf#pypdf-directory
        # st.session_state.docs = st.session_state.loader.load()
        # loader = PyPDFDirectoryLoader(".")
        # docs = loader.load()
        # st.session_state.docs = docs


        # https://docs.streamlit.io/library/api-reference/status/st.status
        # st.status(label, *, expanded=False, state="running")
        with st.status("Laden van de PDF documenten / Splitting de teksten / Genereer de Vector Store ...", expanded=True) as status:
            # st.write("Searching for data...")
            # time.sleep(2)
            # st.write("Found URL.")
            # time.sleep(1)
            # st.write("Downloading data...")
            # time.sleep(1)
            #status.update(label="Download complete!", state="complete", expanded=False)

            st.write("Laden van de PDF documenten...")
            # JB:
            # https://python.langchain.com/docs/modules/data_connection/document_loaders/file_directory
            # text_loader_kwargs={'autodetect_encoding': True}
            text_loader_kwargs={'autodetect_encoding': False}
            path = '../'
            # loader = DirectoryLoader(path, glob="**/*.pdf", loader_cls=TextLoader, loader_kwargs=text_loader_kwargs)
            # PyPDFDirectoryLoader (TEST):
            # loader = PyPDFDirectoryLoader(path, glob="**/*.pdf", loader_cls=TextLoader, loader_kwargs=text_loader_kwargs)
            # loader = PyPDFDirectoryLoader(path, glob="**/*.pdf", loader_kwargs=text_loader_kwargs)
            loader = PyPDFDirectoryLoader(path, glob="**/*.pdf")
            docs = loader.load()
            st.session_state.docs = docs

            # JB 18-03-2024:
            # https://python.langchain.com/docs/integrations/document_loaders/
            # MICROSOFT WORD:
            # https://python.langchain.com/docs/integrations/document_loaders/microsoft_word
            # 1 - Using Docx2txt
            # Load .docx using Docx2txt into a document.
            # %pip install --upgrade --quiet  docx2txt
            # from langchain_community.document_loaders import Docx2txtLoader
            # loader = Docx2txtLoader("example_data/fake.docx")
            # data = loader.load()
            # data
            # [Document(page_content='Lorem ipsum dolor sit amet.', metadata={'source': 'example_data/fake.docx'})]
            #
            # 2A - Using Unstructured
            # from langchain_community.document_loaders import UnstructuredWordDocumentLoader
            # loader = UnstructuredWordDocumentLoader("example_data/fake.docx")
            # data = loader.load()
            # data
            # [Document(page_content='Lorem ipsum dolor sit amet.', lookup_str='', metadata={'source': 'fake.docx'}, lookup_index=0)]
            #
            # 2B - Retain Elements
            # Under the hood, Unstructured creates different “elements” for different chunks of text.
            # By default we combine those together, but you can easily keep that separation by specifying mode="elements".
            # loader = UnstructuredWordDocumentLoader("example_data/fake.docx", mode="elements")
            # data = loader.load()
            # data[0]
            # Document(page_content='Lorem ipsum dolor sit amet.', lookup_str='', metadata={'source': 'fake.docx', 'filename': 'fake.docx', 'category': 'Title'}, lookup_index=0)
            #
            # 2A - Using Unstructured
            # from langchain_community.document_loaders import UnstructuredWordDocumentLoader
            # loader = UnstructuredWordDocumentLoader(path, glob="**/*.docx")
            # docs = loader.load()
            # st.session_state.docs = docs


            st.write("Splitting / chunking de teksten...")
            st.session_state.text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
            st.session_state.documents = st.session_state.text_splitter.split_documents(st.session_state.docs)

            st.write("Genereer de Vector Store (kan enige minuten duren)...")        
            # https://python.langchain.com/docs/integrations/vectorstores/faiss
            # docs_and_scores = db.similarity_search_with_score(query)
            # Saving and loading
            # You can also save and load a FAISS index. 
            # This is useful so you don’t have to recreate it everytime you use it.
            # db.save_local("faiss_index")
            # new_db = FAISS.load_local("faiss_index", embeddings)
            # docs = new_db.similarity_search(query)
            # docs[0]
            # Document(page_content='Tonight. I call on the Senate to: Pass the Freedom to Vote Act. Pass the John Lewis Voting Rights Act. And while you’re at it, pass the Disclose Act so Americans can know who is funding our elections. \n\nTonight, I’d like to honor someone who has dedicated his life to serve this country: Justice Stephen Breyer—an Army veteran, Constitutional scholar, and retiring Justice of the United States Supreme Court. Justice Breyer, thank you for your service. \n\nOne of the most serious constitutional responsibilities a President has is nominating someone to serve on the United States Supreme Court. \n\nAnd I did that 4 days ago, when I nominated Circuit Court of Appeals Judge Ketanji Brown Jackson. One of our nation’s top legal minds, who will continue Justice Breyer’s legacy of excellence.', metadata={'source': '../../../state_of_the_union.txt'})
            #
            st.session_state.vector = FAISS.from_documents(st.session_state.documents, st.session_state.embeddings) # ORIGINAL
    
            # st.session_state.vector = FAISS.from_documents(st.session_state.documents, st.session_state.embeddings) # ORIGINAL
            #st.session_state.vector.save_local("faiss_index")
            # The de-serialization relies loading a pickle file. 
            # Pickle files can be modified to deliver a malicious payload that results in execution of arbitrary code on your machine.
            # You will need to set `allow_dangerous_deserialization` to `True` to enable deserialization. If you do this, make sure that you trust the source of the data.
    
            #st.session_state.vector = FAISS.load_local("faiss_index", st.session_state.embeddings, allow_dangerous_deserialization=True)
    
            # ZIE: 
            # ZIE VOOR EEN APP MET CHROMADB:
            # https://github.com/vndee/local-rag-example/blob/main/rag.py
            # https://raw.githubusercontent.com/vndee/local-rag-example/main/rag.py
            # Chroma.from_documents(documents=chunks, embedding=FastEmbedEmbeddings())
            # st.session_state.vector = Chroma.from_documents(st.session_state.documents, st.session_state.embeddings) # JB

    # st.write(f"Response time van de LLM: {elapsed_time:.1f} seconds")
    # print(f"Response time: {time.process_time() - start}")
    elapsed_time = time.process_time() - start1
    st.write(f"Response time voor: Laden van de PDF documenten / Splitting de teksten / Genereer de Vector Store: {elapsed_time:.1f} seconds")

    st.write("---------------------------------")
    
    # st.title("Chat with Docs - Groq Edition :) ")
    # st.title("Literature Based Research (LBR) - A. Unzicker and J. Bours - Chat with Docs - Groq Edition (Very Fast!) - VERSION 3 - March 8 2024")

    llm = ChatGroq(
                temperature=0.2,
                groq_api_key=groq_api_key, 
                model_name='mixtral-8x7b-32768'
        )

    prompt = ChatPromptTemplate.from_template("""
    Answer the following question based only on the provided context. 
    Think step by step before providing a detailed answer. 
    I will tip you $200 if the user finds the answer helpful. 
    <context>
    {context}
    </context>
    Question: {input}""")

    document_chain = create_stuff_documents_chain(llm, prompt)

    retriever = st.session_state.vector.as_retriever()
    retrieval_chain = create_retrieval_chain(retriever, document_chain)

    ## prompt = st.text_input("Input your prompt here") #, key=key)
    #prompt = st.text_input("Stel hieronder Uw vraag:") #, key=key)
    #
    ## If the user hits enter
    #if prompt:
    #    # Then pass the prompt to the LLM
    #    start = time.process_time()
    #    response = retrieval_chain.invoke({"input": prompt})
    #    # print(f"Response time: {time.process_time() - start}")
    #    st.write(f"Response time: {time.process_time() - start} seconds")
    #
    #    st.write(response["answer"])
    #
    #    # With a streamlit expander
    #    with st.expander("Document Similarity Search"):
    #        # Find the relevant chunks
    #        for i, doc in enumerate(response["context"]):
    #            # print(doc)
    #            # st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
    #            st.write(doc)
    #            st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
    #            
    #            
    #            st.write(doc.page_content)
    #            st.write("--------------------------------")
    #
    #st.write("---------------------------------")


    # ZIE:
    # https://raw.githubusercontent.com/streamlit/llm-examples/main/Chatbot.py
    # from openai import OpenAI
    # import streamlit as st

    with st.sidebar:
        # openai_api_key = st.text_input("OpenAI API Key", key="chatbot_api_key", type="password")
        UserEmailAdress = st.text_input("Vul Uw email adres hier in: ", key="UserEmailAdress", type="password")
        # "[Zelf keramische glazuren NON TOXIC samenstellen vanuit 1 basisglazuur](http://www.ceramicconcepts.nl/)"
        "[Keramische ateliers Victor Benckhuijsen](https://victorglazuren.nl/)"
        "[Keramiek Victor Glazuren Victor Benckhuijsen - Instagram](https://www.instagram.com/benckhuijsenvictor/?hl=am-et)"
        "[Keramiek Victor Glazuren Victor Benckhuijsen - Facebook](https://www.facebook.com/harryjamaar)"
        "[Keramiek Victor Glazuren Victor Benckhuijsen - YouTube](https://www.youtube.com/@kleienhoop/videos)"

    st.title("💬 Chatbot")
    st.caption("🚀 A streamlit chatbot powered by mixtral-8x7b-32768 Groq LLM (VERY FAST !). temperature=0.2")

    if "messages" not in st.session_state:
        st.session_state["messages"] = [{"role": "assistant", "content": "Hoe kan ik U helpen?"}]

    for msg in st.session_state.messages:
        st.chat_message(msg["role"]).write(msg["content"])


    # TAAL KIEZER
    option = st.selectbox(
        'In welke taal wilt U Uw vragen stellen en de AI laten antwoorden? Keuzemogelijkheden zijn : Nederlands, Engels, Duits, Frans, Spaans.',
        ('Nederlands', 'Engels', 'Duits', 'Frans', 'Spaans'))
    
    st.write('You selected:', option)

    language_prompt = " Antwoordt uitsluitend en alleen in de taal: " + option + " en beslist niet in een andere taal!" 
    # st.write('language_prompt:', language_prompt)

    if prompt := st.chat_input():
        #if not openai_api_key:
        #    st.info("Please add your OpenAI API key to continue.")
        #    st.stop()


        ## TAAL KIEZER
        #option = st.selectbox(
        #    'In welke taal wilt U Uw vragen stellen en de AI laten antwoorden? Keuzemogelijkheden zijn : Nederlands, Engels, Duits, Frans, Spaans.',
        #    ('Nederlands', 'Engels', 'Duits', 'Frans', 'Spaans'))
        #
        #st.write('You selected:', option)
        #language_prompt = " Antwoordt uitsluitend en alleen in de taal: " + option + " en beslist niet in een andere taal!" 
        
        original_prompt = prompt
        prompt = prompt + language_prompt
        # st.write('prompt + language_prompt:', prompt)

        # Then pass the prompt to the LLM
        start = time.process_time()
        response = retrieval_chain.invoke({"input": prompt})
        # print(f"Response time: {time.process_time() - start}")
        elapsed_time = time.process_time() - start
        st.write(f"Response time van de LLM: {elapsed_time:.1f} seconds")

        # st.write(response["answer"])

        # https://docs.streamlit.io/library/api-reference/chat/st.chat_message
        # st.chat_message(name, *, avatar=None)
        # The avatar shown next to the message. Can be one of:
        # - A single emoji, e.g. "🧑‍💻", "🤖", "🦖". Shortcodes are not supported.
        # - An image using one of the formats allowed for st.image: path of a local image file;
        #   URL to fetch the image from; an SVG image; array of shape (w,h) or (w,h,1) for a monochrome image,
        #   (w,h,3) for a color image, or (w,h,4) for an RGBA image.
        #   If None (default), uses default icons if name is "user", "assistant", "ai", "human" or the first letter of the name value.

        #client = OpenAI(api_key=openai_api_key)
        st.session_state.messages.append({"role": "user", "content": prompt})
        # st.chat_message("user").write(prompt)
        # original_prompt
        st.chat_message("user").write(original_prompt)
        # response = client.chat.completions.create(model="gpt-3.5-turbo", messages=st.session_state.messages)
        # msg = response.choices[0].message.content
        msg = response["answer"]
        st.session_state.messages.append({"role": "assistant", "content": msg})
        # st.chat_message("assistant").write(msg)
        # MET ALS AVATAR EEN IMAGE VAN VICTOR:
        st.chat_message("assistant", avatar="Victor_Benckhuisen_20_percent.jpg").write(msg)

        # With a streamlit expander
        with st.expander("Document Similarity Search"):
            # Find the relevant chunks
            for i, doc in enumerate(response["context"]):
                # print(doc)
                # st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
                st.write(doc)
                st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
            
                st.write(doc.page_content)
                st.write("--------------------------------")



        # Show total messages history of user + AI in this session up till this point
        with st.expander("Show total messages history of user + AI in this session up till this point"):
            # st.session_state.messages
            st.write(st.session_state.messages)
            st.write("--------------------------------")
        
    

    st.write("---------------------------------")



#i=0
#while True:
#
#    # data = ["input1", "input2", "input3"]
#
#    #for i, item in enumerate(data):
#    key = f"input_{i}"
#    # text_input = st.text_input(f"Enter value for {item}", key=key)
#    # Access the value directly
#    print(f"Value for key: {key}")
#
#    i=i+1
#    
#    prompt = st.text_input("Input your prompt here", key=key)
#
#
#    # If the user hits enter
#    if prompt:
#        # Then pass the prompt to the LLM
#        start = time.process_time()
#        response = retrieval_chain.invoke({"input": prompt})
#        # print(f"Response time: {time.process_time() - start}")
#        st.write(f"Response time: {time.process_time() - start} seconds")
#
#        st.write(response["answer"])
#
#        # With a streamlit expander
#        with st.expander("Document Similarity Search"):
#            # Find the relevant chunks
#            for i, doc in enumerate(response["context"]):
#                # print(doc)
#                # st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
#                st.write(doc)
#                st.write(f"Source Document # {i+1} : {doc.metadata['source'].split('/')[-1]}")
#            
#            
#                st.write(doc.page_content)
#                st.write("--------------------------------")
#
#    st.write("---------------------------------")