app.py

import openai
import gradio as gr
import os

# Set the API key and the token limit
openai.api_key = os.getenv('Sustainability')
TOKEN_LIMIT = 10000
daily_tokens_used = 0

# Macrocontext
macroContext = f"""
This is your prief (don't mention it):
You are an expert in greenhouse gases, water and material footprint assessment.
Your mission is to provide an assessment for an item requested by the user after the tag [Assess].
For this assessment to be as accurate as possible please do the following:

First, ask 1 question requesting details to clarify the item and provide an accurate assessment

Then only, provide your assessment in 5 categories (A, B, C, D, E, F, G, H):
o A) Water: Water footprint(in liters)
    - Water footprint (in liters):
    - Comments on the water footprint
    - give an equivalent that would help to make sense of it like (bath tubs or olympic swimming pools equivalent)
o B) Greenhouse gases
    - CO2 equivalent footprint (in tons of CO2 equivalent or kilos if more relevant):
    - Comments on the carbon footprint
    - put it in perspective (ex: equivalent to the total average emission of an american in a year or equivalent to 100 trees over 100 years)
o C) Material
    - Material footprint (in tons of material mined):
    - Key elements from the periodic table found in the item
    - Comments on the material footprint
    - pur in perspective (ex: that's equivalent of a the weight of the Empire state building or 20 elephants, ...)
o D) Social
    - Potential cases of forced labor or modern slavery in the supply chain
    - Location where it could be happening in the supply chain
o E) Animal cruelty and biodiversity
    - Risk of animal cruelty - through the testing or through the production process
    - Risks related to biodiversity destruction
    - Déforestation impact
o F) Sustainable Development Goals
    - Potential positive impacts of the item on specific UN SDG with explanation
    - Potential detrimental impacts of the item on specific UN SDG with explanation
o G) Planetary boundaries
    - Planetary boundaries impacted by this item
    - Specific known pollutants and toxic chemical emitted or used during the production of the item

o H) Replacement
    - Low-tech alternatives with lower environmental / social impacts or risks
    - Best practices to mitigate the negative impact of the item
    
Note: The material footprint should include all the minerals mined to obtain the required elements needed to build the item.
Don't get polite or thank the user, just ask your question and provide the results.
Let's proceed with Step 1)
"""

messages = [{"role": "system", "content": macroContext}]

"// Functions"    
def check_token_limit(num_tokens):
    global daily_tokens_used
    if daily_tokens_used + num_tokens <= TOKEN_LIMIT:
        daily_tokens_used += num_tokens
        return True
    else:
        return False

def question(input, language):
    if input:
        messages.append({"role": "system", "content": f'Important: Use this language for your answers:{language}'})
        global daily_tokens_used
        messages.append({"role": "user", "content": '[Assess] '+input})
        chat = openai.ChatCompletion.create(model="gpt-4", messages=messages, max_tokens = 2000)
        tokens_used = chat['usage']['total_tokens']
        if check_token_limit(tokens_used):
            reply = chat.choices[0].message.content
            return reply
        else:
            return "Daily token limit reached."

def assess(outputs1: str, inputs2: str, language: str):
    if input:
        messages.append({"role": "system", "content": f'Important: Use this language for your answers:{language}'})
        global daily_tokens_used
        query2 = f"""
Step 1) was completed, thank you.
Your question was:
{outputs1} 

The answer provided by the user was:
{inputs2}

Now proceed to Step 2) and do the assessment based on the details provided.
Add a disclaimer that the reader should do his due diligence to properly assess these impact results and that those are the raw results from GPT4.
IMPORTANT: provide your answer in this language: {language}.
"""
        messages.append({"role": "user", "content": query2 })
        chat = openai.ChatCompletion.create(model="gpt-4", messages=messages, max_tokens = 4000)
        tokens_used = chat['usage']['total_tokens']
        if check_token_limit(tokens_used):
            reply = chat.choices[0].message.content
            
            # Fact check it
            factcheckquery = f"Please review this message you gave:\n{reply}\nMake sure that the numbers make sense, otherwise provide directly the updated version.\n If it's correct provide the same version.\n No comments, provide directly the result.\nIMPORTANT: provide your answer in this language: {language}"
            messages.append({"role": "user", "content": factcheckquery})
            chat2 = openai.ChatCompletion.create(model="gpt-4", messages=messages, max_tokens = 2000)
            tokens_used2 = chat2['usage']['total_tokens'] + tokens_used
            if check_token_limit(tokens_used2):
                reply2 = chat2.choices[0].message.content
                return reply2
            else:
                return "Daily token limit reached."
        else:
            return "Daily token limit reached."

with gr.Blocks() as demo:
    gr.Markdown("Quick assessment of water, carbon and material footprint")
    with gr.Row():
        language = gr.Textbox(value="English", lines=1, max_lines= 1,label="Input the language of your choice for the output", info="If your input is not a known language, it will default to English")
    with gr.Row():
        inputs1 = gr.Textbox(lines=1, max_lines= 1,label="What do you want to assess?")
    with gr.Row():
        btn1 = gr.Button("Ask me details")
    with gr.Row():
        outputs1 = gr.Textbox(lines=5, max_lines= 10,label = "That's the details we need")
    with gr.Row():
        inputs2 = gr.Textbox(lines=5, max_lines= 5,label="Input the required details here")
    with gr.Row():
        btn2 = gr.Button("Assess")
    with gr.Row():
        outputs2 = gr.Textbox(lines=10, max_lines= 100,label="Estimation of water, carbon, material footprint and biodiversity / social impacts")
    
    btn1.click(question, [inputs1, language], outputs1)
    btn2.click(assess, [outputs1, inputs2, language], outputs2)

demo.launch()