app.py

import streamlit as st
from openai import OpenAI
from together import Together
from pymongo.mongo_client import MongoClient
from pymongo.server_api import ServerApi
from datetime import datetime
import time

st.set_page_config(
    page_title="Chat with me!",
    page_icon="🌎",
    initial_sidebar_state="expanded",
    layout="wide"
)
st.markdown(
    """ <style>
            div[role="radiogroup"] >  :first-child{
                display: none !important;
            }
        </style>
        """,
    unsafe_allow_html=True
)

### Setting up the session state 
def generate_tokens(response):
    # for deepseek
    for token in response:
        if hasattr(token, 'choices') and token.choices:
            content = token.choices[0].delta.content
            if content:   
                yield content

def format_personalization(text):
    try:
        for key, value in st.session_state.items():
            text = text.replace(f"[{key.upper()}]", str(value))

    except Exception as e:
        print(text)
        f"Failed to format personalization: {e}"
    return text
    
if 'inserted' not in st.session_state:
    with open('base.txt', 'r') as file:
        st.session_state.base_text = file.read()
    with open('knowledge.txt', 'r') as file:
        st.session_state.knowledge_text = file.read()
    with open('personalization.txt', 'r') as file:
        st.session_state.personalization_text = file.read()

    # web app state
    st.session_state.gotit = False
    st.session_state.submitted = False
    st.session_state.inserted = 0
    st.session_state["model"] = "deepseek/deepseek-chat" #"deepseek/deepseek-r1" #"gpt-4o-2024-08-06" # "o3-mini-2025-01-31" #"deepseek-ai/DeepSeek-V3"
    st.session_state.max_messages = 50
    st.session_state.messages = []
    st.session_state.user_data = {}

    # user info state
    st.session_state.fields = [
    'climate_actions', 'age', 'gender', 'education', 'residence', 'property', 'car',
    'politics', 'impact_open', 'ev', 
    'fossil', 'aerosol', 'diet', 'recycling',
    'user_id'
    ]

    for field in st.session_state.fields:
        st.session_state[field] = ''

    st.session_state.recycling = 0

    st.session_state.start_time = datetime.now()
    st.session_state.convo_start_time = ''

if 'p' not in st.query_params:
    st.query_params['p'] = 't'

def setup_messages():
    # t = personalization, k = knowledge, f = formatting, n = no chat

    if st.query_params["p"] == "f" or st.query_params["p"] == "n":
        st.session_state.system_message = st.session_state.base_text 
    elif st.query_params["p"] == "k":
        st.session_state.system_message = st.session_state.knowledge_text
    elif st.query_params["p"] == "t":
        st.session_state.system_message = format_personalization(st.session_state.personalization_text)

    st.session_state.messages = [{ "role": "system", "content": st.session_state.system_message}]
    st.session_state.convo_start_time = datetime.now()

if st.session_state.model.startswith("deepseek"):
    client = OpenAI( base_url="https://openrouter.ai/api/v1", api_key=st.secrets["OPENROUTER_API_KEY"])
    #client = Together(api_key=st.secrets["TOGETHER_API_KEY"])
else:
    client = OpenAI(api_key=st.secrets["OPENAI_API_KEY"])

### App interface 
with st.sidebar:
    st.markdown("# Let's talk climate action!")
    st.markdown(f"""
    {"☑" if st.session_state.submitted else "☐"} **Step 1. Complete a form.**

    {"☑" if len(st.session_state.messages) > 0 else "☐"} **Step 2. Type in the chat box to start a conversation.**

    🌱 You should ask a climate change related question like:
    - *What are the most effective actions to reduce my carbon emissions?*
    - *What's better for the environment: a year of vegetarianism or skipping one transatlantic flight?*
    - *How do the emissions saved by switching to an EV compare to recycling for a year in terms of trees planted?*

    🧠 If you're unsure about a metric or number, simply ask the chatbot for an explanation.

    ❗ **Do not share any personal information (e.g., name or address) in the chat.**

    ⚠️ You must respond **at least 5 times** before you can submit the conversation. An *End Conversation* button will appear then. You can continue before submitting, but **you must submit the conversation to receive compensation**.
    
    {"☑" if st.session_state.inserted > 1 else "☐"} **Step 3. Use the *End Conversation* button to  submit your response.** 

    ↺ *You can always return to this panel by clicking the arrow on the top left.*
    
    {"🎉 **All done! Please press *Next* in the survey.**" if st.session_state.inserted > 1 else ""}
    """)
    if st.session_state.gotit == False or st.session_state.submitted == False:
        st.session_state.gotit = st.button("Let's start!", key=None, help=None, use_container_width=True) 


@st.dialog('Form')
def form():
    st.markdown("**❗ Please answer every question to proceed.**")
    st.session_state.user_id = st.text_input(label="Enter your Prolific ID", value=st.session_state.user_id)
    st.session_state.age = st.text_input("How old are you in years?")
    st.session_state.gender = st.radio("Do you describe yourself as a man, a woman, or in some other way?", 
                ['','Man', 'Woman', 'Other'])
    st.session_state.education = st.radio("What is the highest level of education you completed?", 
                ['', 
                'Did not graduate high school', 
                'High school graduate, GED, or alternative', 
                'Some college, or associates degree',
                "Bachelor's (college) degree or equivalent",
                "Graduate degree (e.g., Master's degree, MBA)",
                'Doctorate degree (e.g., PhD, MD)'])
    st.session_state.politics = st.radio('What is your political orientation?', ['', 'Extremely liberal', 'Liberal', 'Slightly liberal', 'Moderate', 'Slightly conservative', 'Conservative', 'Extremely conservative'])
    st.session_state.residence = st.radio("What type of a community do you live in?", 
                ['', 'Urban','Suburban','Rural','Other'])
    st.session_state.property = st.radio("Do you own or rent the home in which you live?", 
                ['', 'Own', 'Rent', 'Neither (I live rent-free)', 'Other' ])
    st.session_state.car = st.radio("Do you own or lease a car?",
                ['', 'Own', 'Lease', 'Neither (I do not own or lease a car)'])
    
    st.session_state.climate_actions = st.text_area('Please describe any actions you are taking to address climate change? Write *None* if you are not taking any.')
    st.session_state.impact_open = st.text_area('What do you believe is the single most effective action you can take to reduce carbon emissions that contribute to climate change?')

    st.session_state.recycling = st.slider('What percentage of plastic produced gets recycled?', 0, 100, value=0)

    st.markdown("**Do you agree or disagree with the following statements?**")
    st.session_state.ev = st.radio("Electric vehicles don't have enough range to handle daily travel demands.", ["", "Strongly Disagree", "Disagree", "Neutral", "Agree", "Strongly Agree"])
    st.session_state.fossil = st.radio('The fossil fuel industry is trying to shift the blame away from themselves by emphasizing the importance of individual climate action.', ["", "Strongly Disagree", "Disagree", "Neutral", "Agree", "Strongly Agree"])
    st.session_state.aerosol = st.radio('The use of aerosol spray cans is a major cause of climate change.', ["", "Strongly Disagree", "Disagree", "Neutral", "Agree", "Strongly Agree"])
    st.session_state.diet = st.radio('Lab-grown meat produces up to 25 times more CO2 than real meat.', ["", "Strongly Disagree", "Disagree", "Neutral", "Agree", "Strongly Agree"])


    columns_form = st.columns((1,1,1))
    with columns_form[2]:
        submitted = st.button("Proceed",use_container_width=True, 
                              help = 'Make sure you answer every question',
                              disabled = not (all(st.session_state[field] != '' for field in st.session_state.fields) and st.session_state.recycling != 0))

    if submitted:

        st.session_state.user_data = {key: st.session_state[key] for key in st.session_state.fields}
        st.session_state.user_data["model"] = st.session_state["model"]
        st.session_state.user_data["condition"] = st.query_params['p']
        st.session_state.user_data["start_time"] = st.session_state.start_time
        st.session_state.user_data["inserted"] = st.session_state.inserted
        st.session_state.user_data["submission_time"] = datetime.now()
        
        with MongoClient(st.secrets["mongo"],server_api=ServerApi('1')) as client:
                db = client.chat
                collection = db.app
                user_data = st.session_state.user_data
                collection.insert_one(user_data)  
                st.session_state.inserted += 1
        st.session_state.submitted = True
        setup_messages()
        st.rerun()

if st.session_state.gotit and st.session_state.submitted == False:
    form()

for message in st.session_state.messages:
    if message['role']!='system':
        with st.chat_message(message["role"]):
            st.markdown(message["content"])

@st.dialog('Submit conversation')
def submit():
    st.markdown("You must answer all questions marked with a ❗ to submit.")
    if st.query_params["p"] != "n":
        st.slider('❗ How would you rate the conversation on a scale from *Terrible* to *Perfect*?', 0, 100, format="", key="score", value=50)
        st.slider('❗ How personalized did the conversation feel, on a scale from *Not at all* to *Extremely personalized*?', 0, 100, format="", key="personalization_score", value=50)
        st.slider('❗ How knowledgeable do you feel the chatbot was, on a scale from *Not at all* to *Extremely knowledgeable*?', 0, 100, format="", key="knowledge_score", value=50)
    else:
        st.session_state.score = 0
        st.session_state.personalization_score = 0
        st.session_state.knowledge_score = 0

    st.text_area('Any feedback?',key="feedback")
    if st.button('Submit', key=None, help=None, use_container_width=True, disabled=st.session_state.score==50 or st.session_state.personalization_score==50):
        keys = ["score", "personalization_score", "knowledge_score", 
                "feedback", 
                "inserted", "messages", "convo_start_time"]

        st.session_state.user_data.update({key: st.session_state[key] for key in keys})
        
        st.session_state.user_data["convo_end_time"] = datetime.now()

        with MongoClient(st.secrets["mongo"],server_api=ServerApi('1')) as client:
                db = client.chat
                collection = db.app
                user_data = st.session_state.user_data
                del user_data['_id']

                collection.insert_one(user_data)  
                st.session_state.inserted += 1
                
                st.success('**Your conversation has been submitted! Please proceed with the survey.**', icon="✅")

                time.sleep(10)
                setup_messages()
                st.rerun()

if len(st.session_state.messages) >= st.session_state.max_messages:
    st.info(
        "You have reached the limit of messages for this conversation. Please end and submit the conversatione."
    )

elif st.session_state.submitted == False:
    pass

elif st.query_params["p"] == "n":
    st.markdown("""
                You have not been selected to have a conversation with the chatbot.

                ❗ **Please press *End Conversation* to submit your data and proceed with the survey. You have to submit to receive compensation.** 
                """)
    columns = st.columns((1,1,1))
    with columns[2]:
        if st.button("End Conversation",use_container_width=True):
            submit()

elif prompt := st.chat_input("Ask a question about climate action..."):   

    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)

    with st.chat_message("assistant"):
        try:
            stream = client.chat.completions.create(
                model=st.session_state["model"],
                messages=[
                    {"role": m["role"], "content": m["content"]}
                    for m in st.session_state.messages
                ],
                stream=True
            )
            
            if False: #st.session_state.model.startswith("deepseek"):
                
                response = st.write_stream(generate_tokens(stream)) 
            else:
                response = st.write_stream(stream)

            st.session_state.messages.append(
                {"role": "assistant", "content": response}
            )
        except:
            rate_limit_message = """
                An error has occured or you've reached the maximum conversation length. Please submit the conversation.
            """
            st.session_state.messages.append(
                {"role": "assistant", "content": rate_limit_message}
            )
            st.session_state.max_messages = len(st.session_state.messages)
            st.rerun()

if len(st.session_state.messages) > 10 or st.session_state.max_messages == len(st.session_state.messages):
    columns = st.columns((1,1,1))
    with columns[2]:
        if st.button("End Conversation",use_container_width=True):
            submit()