app.py

import streamlit as st
from openai import OpenAI
import json, os

#Enable for testing
debug_mode = True

# Initialize OpenAI client
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))  # Replace with your actual API key

# Initialize assistants and vector stores
# (This part should be done outside the main Streamlit app for performance reasons)
# ... (code to initialize assistants and vector stores) ...
#Processing Level
assistant1 = client.beta.assistants.create(
  name="Processing Level",
  instructions="You are an expert dietician. Use you knowledge base to answer questions about the processing level of food product.",
  model="gpt-4o",
  tools=[{"type": "file_search"}],
  temperature=0,
  top_p = 0.85
  )

#Harmful Ingredients
assistant2 = client.beta.assistants.create(
  name="Harmful Ingredients",
  instructions="You are an expert dietician. Use you knowledge base to answer questions about the ingredients in food product.",
  model="gpt-4o",
  tools=[{"type": "file_search"}],
  temperature=0,
  top_p = 0.85
  )

#Harmful Ingredients
assistant3 = client.beta.assistants.create(
  name="Misleading Claims",
  instructions="You are an expert dietician. Use you knowledge base to answer questions about the misleading claims about food product.",
  model="gpt-4o",
  tools=[{"type": "file_search"}],
  temperature=0,
  top_p = 0.85
  )

# Create a vector store
vector_store1 = client.beta.vector_stores.create(name="Processing Level Vec")

# Ready the files for upload to OpenAI
file_paths = ["Processing_Level.docx"]
file_streams = [open(path, "rb") for path in file_paths]

# Use the upload and poll SDK helper to upload the files, add them to the vector store,
# and poll the status of the file batch for completion.
file_batch1 = client.beta.vector_stores.file_batches.upload_and_poll(
  vector_store_id=vector_store1.id, files=file_streams
)

# You can print the status and the file counts of the batch to see the result of this operation.
print(file_batch1.status)
print(file_batch1.file_counts)

# Create a vector store
vector_store2 = client.beta.vector_stores.create(name="Harmful Ingredients Vec")

# Ready the files for upload to OpenAI
file_paths = ["Ingredients.docx"]
file_streams = [open(path, "rb") for path in file_paths]

# Use the upload and poll SDK helper to upload the files, add them to the vector store,
# and poll the status of the file batch for completion.
file_batch2 = client.beta.vector_stores.file_batches.upload_and_poll(
  vector_store_id=vector_store2.id, files=file_streams
)

# You can print the status and the file counts of the batch to see the result of this operation.
print(file_batch2.status)
print(file_batch2.file_counts)

# Create a vector store
vector_store3 = client.beta.vector_stores.create(name="Misleading Claims Vec")

# Ready the files for upload to OpenAI
file_paths = ["MisLeading_Claims.docx"]
file_streams = [open(path, "rb") for path in file_paths]

# Use the upload and poll SDK helper to upload the files, add them to the vector store,
# and poll the status of the file batch for completion.
file_batch3 = client.beta.vector_stores.file_batches.upload_and_poll(
  vector_store_id=vector_store3.id, files=file_streams
)

# You can print the status and the file counts of the batch to see the result of this operation.
print(file_batch3.status)
print(file_batch3.file_counts)

#Processing Level
assistant1 = client.beta.assistants.update(
  assistant_id=assistant1.id,
  tool_resources={"file_search": {"vector_store_ids": [vector_store1.id]}},
)

#harmful Ingredients
assistant2 = client.beta.assistants.update(
  assistant_id=assistant2.id,
  tool_resources={"file_search": {"vector_store_ids": [vector_store2.id]}},
)

#Misleading Claims
assistant3 = client.beta.assistants.update(
  assistant_id=assistant3.id,
  tool_resources={"file_search": {"vector_store_ids": [vector_store3.id]}},
)

def analyze_processing_level(ingredients, product_info, assistant_id):
    global debug_mode
    thread = client.beta.threads.create(
        messages=[
            {
                "role": "user",
                "content": "Categorize product " + product_info["brandName"] + product_info["productName"] + " that has following ingredients: " + ', '.join(ingredients) + " into Group A, Group B, or Group C based on the document. The output must only be the group category name (Group A, Group B, or Group C) alongwith the definition of the respective category. If the group category cannot be determined, output 'NOT FOUND'.",
            }
        ]
    )
    
    run = client.beta.threads.runs.create_and_poll(
        thread_id=thread.id,
        assistant_id=assistant_id,  # Replace with actual assistant ID
        include=["step_details.tool_calls[*].file_search.results[*].content"]
    )
    
    messages = list(client.beta.threads.messages.list(thread_id=thread.id, run_id=run.id))

    message_content = messages[0].content[0].text
    annotations = message_content.annotations
    #citations = []
    for index, annotation in enumerate(annotations):
        message_content.value = message_content.value.replace(annotation.text, "")
        #if file_citation := getattr(annotation, "file_citation", None):
        #    cited_file = client.files.retrieve(file_citation.file_id)
        #    citations.append(f"[{index}] {cited_file.filename}")

    if debug_mode:
        print(message_content.value)
    processing_level_str = message_content.value
    return processing_level_str

def analyze_harmful_ingredients(ingredients, product_info, assistant_id):
    global debug_mode
    thread = client.beta.threads.create(
        messages=[
            {
                "role": "user",
                "content": "Provide detailed information about product " + product_info["brandName"] + product_info["productName"] + " that has following ingredients: " + ', '.join(ingredients) + ". The output must be in JSON format: {<ingredient_name>: <information from the document>}. If information about an ingredient is not found in the documents, the value for that ingredient must start with the prefix '(NOT FOUND IN DOCUMENT)' followed by the LLM's response based on its own knowledge.",
            }
        ]
    )
    
    run = client.beta.threads.runs.create_and_poll(
        thread_id=thread.id,
        assistant_id=assistant_id,  # Replace with actual assistant ID
        include=["step_details.tool_calls[*].file_search.results[*].content"]
    )
    
    messages = list(client.beta.threads.messages.list(thread_id=thread.id, run_id=run.id))
    message_content = messages[0].content[0].text
    annotations = message_content.annotations

    #citations = []

    #print(f"Length of annotations is {len(annotations)}")

    for index, annotation in enumerate(annotations):
      if file_citation := getattr(annotation, "file_citation", None):
          #cited_file = client.files.retrieve(file_citation.file_id)
          #citations.append(f"[{index}] {cited_file.filename}")
          message_content.value = message_content.value.replace(annotation.text, "")
  
    if debug_mode:
      ingredients_not_found_in_doc = []
      print(message_content.value)
      for key, value in json.loads(message_content.value.replace("```", "").replace("json", "")).items():
          if value.startswith("(NOT FOUND IN DOCUMENT)"):
              ingredients_not_found_in_doc.append(key)
      print(f"Ingredients not found in the harmful ingredients doc are {','.join(ingredients_not_found_in_doc)}")
    harmful_ingredient_analysis = json.loads(message_content.value.replace("```", "").replace("json", "").replace("(NOT FOUND IN DOCUMENT) ", ""))
    
    harmful_ingredient_analysis_str = ""
    for key, value in harmful_ingredient_analysis.items():
      harmful_ingredient_analysis_str += f"{key}: {value}\n"
    return harmful_ingredient_analysis_str

def analyze_claims(claims, assistant_id):
    global debug_mode
    thread = client.beta.threads.create(
        messages=[
            {
                "role": "user",
                "content": "Provide detailed information about the following claims: " + ', '.join(claims) + ". The output must be in JSON format: {<claim_name>: <information from the document>}. If information about a claim is not found in the documents, the value for that claim must start with the prefix '(NOT FOUND IN DOCUMENT)' followed by the LLM's response based on its own knowledge.",
            }
        ]
    )
    
    run = client.beta.threads.runs.create_and_poll(
        thread_id=thread.id,
        assistant_id=assistant_id,  # Replace with actual assistant ID
        include=["step_details.tool_calls[*].file_search.results[*].content"]
    )
    
    messages = list(client.beta.threads.messages.list(thread_id=thread.id, run_id=run.id))

    message_content = messages[0].content[0].text
    
      
    annotations = message_content.annotations
    
    #citations = []
    
    #print(f"Length of annotations is {len(annotations)}")
    
    for index, annotation in enumerate(annotations):
          if file_citation := getattr(annotation, "file_citation", None):
              #cited_file = client.files.retrieve(file_citation.file_id)
              #citations.append(f"[{index}] {cited_file.filename}")
              message_content.value = message_content.value.replace(annotation.text, "")
      
    if debug_mode:
        claims_not_found_in_doc = []
        print(message_content.value)
        for key, value in json.loads(message_content.value.replace("```", "").replace("json", "")).items():
              if value.startswith("(NOT FOUND IN DOCUMENT)"):
                  claims_not_found_in_doc.append(key)
        print(f"Claims not found in the doc are {','.join(claims_not_found_in_doc)}")
    claims_analysis = json.loads(message_content.value.replace("```", "").replace("json", "").replace("(NOT FOUND IN DOCUMENT) ", ""))

    claims_analysis_str = ""
    for key, value in claims_analysis.items():
      claims_analysis_str += f"{key}: {value}\n"
    
    return claims_analysis_str

def generate_final_analysis(product_info, processing_level, harmful_ingredient_analysis, claims_analysis):
    global debug_mode
    system_prompt = """You are provided with a detailed analysis of a food product. Your task is to generate actionable insights to help the user decide whether to consume the product, at what frequency, and identify any potential harms or benefits. Consider the context of consumption to ensure the advice is personalized and practical.

Use the following criteria to generate your response:

1. **Nutrition Analysis:**
- How processed is the product?

2. **Harmful Ingredients:**
- Identify any harmful or questionable ingredients.

3. **Misleading Claims:**
- Are there any misleading claims made by the brand?

Additionally, consider the following while generating insights:

1. **Consumption Context:**
- Is the product being consumed for health reasons or as a treat?
- Could the consumer be overlooking hidden harms?
- If the product is something they could consume daily, should they?
- If they are consuming it daily, what potential harm are they not noticing?
- If the product is intended for health purposes, are there concerns the user might miss?

**Output:**
- Recommend whether the product should be consumed or avoided.
- If recommended, specify the appropriate frequency and intended functionality (e.g., treat vs. health).
- Highlight any risks or benefits at that level of consumption."""

    user_prompt = f"""
Product Name: {product_info['brandName']} {product_info['productName']}

Processing Level:
{processing_level}

Ingredient Analysis:
{harmful_ingredient_analysis}

Claims Analysis:
{claims_analysis}
"""
    if debug_mode:
        print(f"\nuser_prompt : \n {user_prompt}")
        
    completion = client.chat.completions.create(
        model="gpt-4o",  # Make sure to use an appropriate model
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt}
        ]
    )

    return completion.choices[0].message.content

# Streamlit app
def main():
    st.title("Food Product Analysis")

    # Create a form for all inputs
    with st.form("product_analysis_form"):
        st.write("Please enter the following information about the product:")
        brand_name = st.text_input("Brand Name")
        product_name = st.text_input("Product Name")
        ingredients = st.text_area("Ingredients (separated by commas)")
        claims = st.text_area("Product Claims (separated by commas)")

        submitted = st.form_submit_button("Analyze Product")

    if submitted:
        # Process inputs
        ingredients_list = []
        claims_list = []
        if ingredients:
            ingredients_list = [ingredient.strip() for ingredient in ingredients.split(',')]
        if claims:
            claims_list = [claim.strip() for claim in claims.split(',')]
        product_info = {
            "brandName": brand_name,
            "productName": product_name
        }

        # Display a message while analyzing
        with st.spinner("Analyzing the product... This may take a moment."):
            # Perform analysis
            processing_level = ""
            harmful_ingredient_analysis = ""
            if ingredients:
                processing_level = analyze_processing_level(ingredients_list, product_info, assistant1.id)
                harmful_ingredient_analysis = analyze_harmful_ingredients(ingredients_list, product_info, assistant2.id)
            else:
                print("No ingredients specified by the user!")
                
            claims_analysis = ""
            if claims:
                claims_analysis = analyze_claims(claims_list, assistant3.id)
            else:
                print("No claims specified by the user!")
            
            # Generate final analysis
            final_analysis = generate_final_analysis(
                product_info,
                processing_level,
                harmful_ingredient_analysis,
                claims_analysis
            )

        # Display results
        st.success("Analysis complete!")
        st.subheader("Final Analysis:")
        st.write(final_analysis)

        # Option to start over
        if st.button("Analyze Another Product"):
            st.rerun()

if __name__ == "__main__":
    main()