Updated

app.py

@@ -3,7 +3,7 @@ import streamlit as st
 st.session_state.em = "0"
 
 import os
-
+import json
 
 
 import requests
@@ -17,19 +17,17 @@ import numpy as np
 st.set_page_config(layout="wide")
 
 
+# Path to the image
+image_path = 'fire.jpg'
 
+# Display the image with st.image
+st.image(image_path, caption='', use_column_width=True)
 
 
-
-started = 'started' in st.session_state
-
-
+started = 'docs' in st.session_state
+exec(open('start2.py').read()) 
 
 
-if started:
-    retriever = st.session_state.retriever
-    rag_chain = st.session_state.rag_chain    
-
 
 os.environ["OPENAI_API_KEY"]  = os.getenv('openkey')
 
@@ -92,7 +90,7 @@ def strip_repeated_dots_and_blanks(text):
 
 
 # Title of the page
-st.title('Question and Answer App')
+st.title('Peerstreet Question and Answer App')
 
 # Text input for the question
 question = st.text_input("Type your question here:")
@@ -100,8 +98,33 @@ question = st.text_input("Type your question here:")
 # A button to submit the question
 submit_button = st.button('Submit')
 
+
+# Create tabs
+Awnser_tab, Content_tab, Info_tab = st.tabs(["Awnser", "Content used to create answer", "Infrmation about this app"])
+
+
+
 # Placeholder for displaying the answer
-answer_placeholder = st.empty()
+with Awnser_tab:
+    answer_placeholder = st.empty()
+with Content_tab:
+    content_placeholder = st.empty()
+with Info_tab:
+    st.markdown("""## Use at your own risk, accuracy of responses are not guaranteed. 
+
+This app base its anwsers on 110 documents filed by the court. This does not include any scanned documents at this point 
+as it takes more work to retrieve the text from them. It does include most orders filed by the court up to Feb 29th.
+
+
+This is a simple RAG (retrieval augmented generation) system and does not consider order of events when 
+retrieving onformation and generating responses. It can also easily missinterpret information, but information used to generate the 
+response is presented in the content tab with link to the full document so that you can read the details in its proper context. 
+
+                
+
+""" 
+                )
+            
 
 # Logic to display an answer when the submit button is pressed
 if submit_button:
@@ -109,13 +132,66 @@ if submit_button:
         # Process the question here (a placeholder answer is used in this example)
         try:
             if started: 
-                Awnser = rag_chain.invoke(question)
-                contexts = retriever.get_relevant_documents(question)
-                answer_placeholder.text(Awnser)  # Display the answer
+                #Awnser = rag_chain.invoke(question)
+                #contexts = retriever.get_relevant_documents(question)
+                answer, selected_items,selected_sources,selected_chunks,highest_simularities = ask(question)
+                answer_placeholder.markdown(answer)  # Display the answer
+
+                    # Prepare the data to be saved
+                
+
+                data_to_save = {
+                    "query": question,
+                    "answer": answer,
+                    "selected_items": selected_items,
+                    "selected_sources": selected_sources,
+                    "selected_chunks": selected_chunks,
+                    "highest_similarities": [f"{sim:.2f}" for sim in highest_simularities]
+                }
+                
+                # The file to which the data will be appended
+                file_path = 'results.json'
+                
+                try:
+                    # Read the existing content of the file
+                    with open(file_path, 'r') as file:
+                        existing_data = json.load(file)
+                except (FileNotFoundError, json.JSONDecodeError):
+                    # If the file doesn't exist or is empty, start with an empty list
+                    existing_data = []
+                
+                # Append the new data
+                existing_data.append(data_to_save)
+                
+                # Write the updated data back to the file
+                with open(file_path, 'w') as file:
+                    json.dump(existing_data, file, indent=4)
+ 
+
+
+
+                url = 'https://cases.stretto.com/public/x247/12208/PLEADINGS/'
+
+
+
+
+
+                string = ""
+                for k in range(len(selected_items)):
+                    temp = " [" +  selected_sources[k] + "](" + url  +  selected_sources[k] +   ")" +  "  text block: " + selected_chunks[k]   + "   Relevance: " +f"{highest_simularities[k]:.2f}"
+
+
+                    string = string + "### Paragraph used. \n" + selected_items[k]  + "\n\n  source:"  + temp  + "\n"
+
+
+                content_placeholder.markdown(string) 
+
+
+
             else:
-                answer_placeholder.text("Waiting for system to wake up   "+     st.session_state.ln + "    " + st.session_state.em )    
+                answer_placeholder.markdown("Waiting for system to wake up   "+     st.session_state.ln + "    " + st.session_state.em )    
         except Exception as e:
-            answer_placeholder.text(e)  # Display the answer
+            answer_placeholder.markdown(e)  # Display the answer
          
     else:
         answer_placeholder.warning("Please type a question.")
@@ -123,10 +199,12 @@ if submit_button:
 
 
     
-if 'retriever' not in st.session_state:
-    st.session_state.em = "mm"
+#if 'retriever' not in st.session_state:
+#    st.session_state.em = "mm"
 
-if 'retriever' not in st.session_state:
-    st.session_state.em = "1"
-    exec(open('start.py').read())
-    st.session_state.em = "2"
+#if 'retriever' not in st.session_state:
+#    st.session_state.em = "1"
+#    exec(open('start.py').read())
+#    st.session_state.em = "2"
+        
+     

embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e59b70d53bc2779e24f76bcc2377fd60b9d3cdabf20b26cd8cfc176ec316292
+size 66072704

start2.py

@@ -0,0 +1,173 @@
+
+from openai import OpenAI
+import numpy as np
+
+from langchain_openai import ChatOpenAI
+from langchain.prompts import ChatPromptTemplate
+from langchain.schema.runnable import RunnablePassthrough
+from langchain.schema.output_parser import StrOutputParser
+from scipy.spatial.distance import cosine
+
+if not started:
+    print("------------starting------------")
+
+
+
+
+    import pickle
+
+
+    # Path to the pickle file where you want to save your data
+    pickle_file_path = 'vectorstore2.pkl'
+
+
+    with open(pickle_file_path, 'rb') as file:
+        st.session_state.docs = pickle.load(file)
+
+
+    st.session_state.embeddings = np.load('embeddings.npy')
+
+
+
+
+
+
+
+def strip_repeated_dots_and_blanks(text):
+    # Replace multiple dots with a single dot
+    text = re.sub(r'\.{2,}', '.', text)
+    # Replace multiple spaces with a single space
+    text = re.sub(r' {2,}', ' ', text)
+    text = re.sub('\n \n', '\n\n', text)
+    
+    return text
+
+
+# Function to get embeddings from OpenAI API
+def get_embeddings(texts):
+    client = OpenAI()
+    embeddings  = []
+    for k in texts: 
+        response = client.embeddings.create(
+            input = k,
+            model="text-embedding-3-small"
+        )
+        embeddings  = embeddings  + [response.data[0].embedding]
+
+    return embeddings  #[item['embedding'] for item in response['data']]
+
+
+
+def cosine_similarity(vec_a, vec_b):
+    # Note: Cosine similarity is 1 - cosine distance
+    return 1 - cosine(vec_a, vec_b)
+
+
+
+
+def askq(query):
+    embeddings = st.session_state.embeddings
+    docs = st.session_state.docs
+
+    question = strip_repeated_dots_and_blanks(query)
+    query_embedding = get_embeddings([query])[0]
+ 
+
+    # Calculate similarity of each text to the query
+    similarities = [cosine_similarity(embedding, query_embedding) for embedding in embeddings]
+
+
+
+    similarities_array = np.array(similarities)
+    highest_indices = np.argpartition(similarities_array, -5)[-5:]
+
+    # Sort these two indices by their similarity values so the highest comes first
+    highest_indices = highest_indices[np.argsort(similarities_array[highest_indices])[::-1]]
+
+#    highest_indices = highest_indices[highest_indices>0.5]
+   # highest_indices = [x for x in highest_indices if x > 0.5]
+#    highest_indices = [index for index in highest_indices if similarities_array[index] > 0.5]
+
+    # Filter indices by their corresponding similarity values
+    filtered_indices_and_values = [(index, similarities_array[index]) for index in highest_indices if similarities_array[index] > 0.5]
+
+    highest_indices = [item[0] for item in filtered_indices_and_values]
+    highest_simularities = [item[1] for item in filtered_indices_and_values]
+
+
+
+    selected_items = [docs[i].page_content for i in highest_indices]
+    selected_sources = [docs[i].metadata['source'] for i in highest_indices]
+    selected_chunks = [docs[i].metadata['chunk'] for i in highest_indices]
+    selected_chunks = [docs[i].metadata['chunk'] for i in similarities_array]
+
+
+    content = ' '.join(selected_items)
+
+
+    # Combine content and question into a single prompt
+    prompt = f"""Anwser question provided given content. If an awnser can't be found in the provided ontent respond that you could not find 
+    the awnser to the question, applogice and say that you will instead tell a lawyer joke follow that with the joke.   
+    Content: {content}\n\nQuestion: {question}\nAnswer:"""
+
+
+    # Define LLM
+    llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0.1)
+    Awnser = llm.invoke(prompt).content
+
+    return Awnser, selected_items,selected_sources,selected_chunks,highest_simularities
+
+
+import numpy as np
+import streamlit as st
+
+# Assuming `strip_repeated_dots_and_blanks`, `get_embeddings`, and `cosine_similarity` are defined elsewhere correctly
+# Assuming `ChatOpenAI` is a correctly defined or imported class for handling OpenAI chat
+
+def ask(query):
+    embeddings = st.session_state.embeddings
+    docs = st.session_state.docs
+
+    question = strip_repeated_dots_and_blanks(query)
+    query_embedding = get_embeddings([query])[0]
+ 
+    # Calculate similarity of each text to the query
+    similarities = [cosine_similarity(embedding, query_embedding) for embedding in embeddings]
+
+    # Create a NumPy array of similarities
+    similarities_array = np.array(similarities)
+    # Get indices of the top 5 most similar texts
+    highest_indices = np.argpartition(similarities_array, -5)[-5:]
+    # Sort the top 5 indices by their similarity values in descending order
+    highest_indices = highest_indices[np.argsort(similarities_array[highest_indices])[::-1]]
+
+    # Filter indices and their corresponding similarity values for those above 0.5
+    filtered_indices_and_values = [(index, similarities_array[index]) for index in highest_indices if similarities_array[index] > 0.4]
+    
+    # Extract filtered indices and their similarities
+    highest_indices = [item[0] for item in filtered_indices_and_values]
+    highest_simularities = [item[1] for item in filtered_indices_and_values]
+
+    # Select items based on filtered indices
+    selected_items = [docs[i].page_content for i in highest_indices]
+    selected_sources = [docs[i].metadata['source'] for i in highest_indices]
+    selected_chunks = [docs[i].metadata['chunk'] for i in highest_indices]
+
+    # Combine selected items into a single content string
+    content = ' '.join(selected_items)
+
+    # Prepare the prompt
+    prompt = f"""Answer the question provided given the content. If an answer can't be found in the provided content, 
+    respond that you could not find the answer to the question, apologize and instead provide a suggestion for where to search for more information related to teh question.
+
+    -------------------   
+    Content: {content}\n\nQuestion: {question}\nAnswer:
+    -------------------
+    
+    """
+
+    # Initialize the LLM (assuming correct implementation or import)
+    llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0.1)
+    answer = llm.invoke(prompt).content
+
+    return answer, selected_items, selected_sources, selected_chunks, highest_simularities

vectorstore2.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eeb601bfdd128945a52712a20a89f9bfd89c85ea1d25215d552f68ca094b012
+size 5582531