initial conversation

app.py

@@ -10,6 +10,8 @@ import streamlit as st
 from transformers import AutoTokenizer
 from sentence_transformers import SentenceTransformer
 
+import streamlit.components.v1 as components
+
 import utils 
 
 PINECONE_KEY = st.secrets["PINECONE_API_KEY"]  # app.pinecone.io
@@ -43,6 +45,7 @@ def init_models():
     return retriever, tokenizer#, vectorstore
 
 retriever, tokenizer = init_models()
+#st.session_state.messages = [{"role":"system", "content":"You are an assistant who helps users find startups to invest in."}]
 
 
 def card(company_id, name, description, score, data_type, region, country, metadata, is_debug):
@@ -59,41 +62,31 @@ def card(company_id, name, description, score, data_type, region, country, metad
         except Exception as e:
             print(f"An error occurred: {str(e)}")
             
-
-        
-    
     markdown = f"""
         <div class="row align-items-start" style="padding-bottom:10px;">
              <div  class="col-md-8 col-sm-8">
                  <b>{name} (<a href='https://{company_id}'>website</a>).</b>
-                 <p style="">
-                    {description}
-                 </p>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>{country}</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>{customer_problem}</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>{business_model}</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <button type='button' onclick="like_company({company_id});">Like</button>
-                    <button type='button' onclick="dislike_company({company_id});">DisLike</button>
+                 <p style="">{description}</p>
              </div>
+             <div  class="col-md-1 col-sm-1"><span>{country}</span></div>
+             <div  class="col-md-1 col-sm-1"><span>{customer_problem}</span></div>
+             <div  class="col-md-1 col-sm-1"><span>{business_model}</span></div>
     """             
 
     if is_debug:
         markdown = markdown + f"""
+        <div  class="col-md-1 col-sm-1" style="display:none;">
+            <button type='button' onclick="like_company({company_id});">Like</button>
+            <button type='button' onclick="dislike_company({company_id});">DisLike</button>
+        </div>
         <div  class="col-md-1 col-sm-1">
             <span>{data_type}</span>
             <span>[Score: {score}</span>
         </div>        
         """             
     markdown = markdown + "</div>"
-    return st.markdown(markdown, unsafe_allow_html=True)
+    #print(f" markdown for {company_id}\n{markdown}")
+    return markdown
 
 
 def index_query(xq, top_k, regions=[], countries=[], index_namespace="websummarized"):
@@ -114,32 +107,8 @@ def index_query(xq, top_k, regions=[], countries=[], index_namespace="websummari
     #xc = st.session_state.index.query(xq, top_k=top_k, include_metadata=True, include_vectors = True)
     return xc
 
-def call_openai(prompt, engine="gpt-3.5-turbo", temp=0, top_p=1.0, max_tokens=4048):
-    try:
-        response = openai.ChatCompletion.create(
-            model=engine,
-            messages=[{"role": "user", "content": prompt}],
-            temperature=temp,
-            max_tokens=max_tokens
-        )
-        print(response)
-        text = response.choices[0].message["content"].strip()
-        return text
-    except openai.error.OpenAIError as e:
-        print(f"An error occurred: {str(e)}")
-    return "Failed to generate a response."
-
-def on_prompt_selected():
-    title = st.session_state.advanced_prompts_select
-    new_prompt = utils.get_prompt(title)
-    if len(new_prompt)>0 and len(new_prompt[0])>0:
-        print(f"Got a prompt for title {title}\n {new_prompt[0]}")
-        st.session_state.prompt_title_editable = st.session_state.advanced_prompts_select
-        st.session_state.advanced_prompt_content = new_prompt[0]
-    else:
-        print(f"No results for title {st.session_state.advanced_prompts_select}")
         
-def run_query(query, prompt, scrape_boost, top_k , regions, countries, is_debug, index_namespace):
+def run_query(query, prompt, scrape_boost, top_k , regions, countries, is_debug, index_namespace, openai_model):
     xq = retriever.encode([query]).tolist()
     try:
         xc = index_query(xq, top_k, regions, countries)
@@ -182,44 +151,61 @@ def run_query(query, prompt, scrape_boost, top_k , regions, countries, is_debug,
     # Create a summarized report focusing on the top3 companies.
     # For every company find its uniqueness over the other companies. Use only information from the descriptions.
     # """
-    prompt_txt = prompt + """        
-    Company descriptions: {descriptions}
-    User query: {query} 
-    """    
-    prompt_template = PromptTemplate(template=prompt_txt, input_variables=["descriptions", "query"])
-    descriptions = str([f"{res['name']}: {res['data']['Summary']}" for res in results[:20] if 'Summary' in res['data']])
-    ntokens = len(descriptions.split(" "))
-    print(f"#Tokens {ntokens}:\n {descriptions}")    
-    prompt = prompt_template.format(descriptions = descriptions, query = query)
-    m_text = call_openai(prompt, engine="gpt-3.5-turbo-16k", temp=0, top_p=1.0)
-
-    m_text
+    if prompt!="":
+        descriptions = "\n".join([f"Description of company \"{res['name']}\":  {res['data']['Summary']}.\n" for res in results[:20] if 'Summary' in res['data']])
+        ntokens = len(descriptions.split(" "))
+
+        print(f"Descriptions ({ntokens} tokens):\n {descriptions[:1000]}")    
+
+        prompt_txt = prompt + """        
+        User query: {query} 
+        Company descriptions: {descriptions}
+        """    
+        prompt_template = PromptTemplate(template=prompt_txt, input_variables=["descriptions", "query"])
+        prompt = prompt_template.format(descriptions = descriptions, query = query)
+
+        print(f"==============================\nPrompt:\n{prompt}\n==============================\n")
+        new_message = {"role": "user", "content": prompt}     
+        m_text = utils.call_openai(prompt, engine=openai_model, temp=0, top_p=1.0)
+
+        m_text
+
+    else:
+        new_message = {"role": "user", "content": query}     
+        
+    st.session_state.messages.append(new_message)
+    render_history()
+    # for message in st.session_state.messages:
+    #     with st.chat_message(message["role"]):
+    #         st.markdown(message["content"])
+#    print(f"History: \n {st.session_state.messages}")
     
     sorted_results = sorted(results, key=lambda x: x['score'], reverse=True)
 
-    st.markdown("<h2>Related companies</h2>", unsafe_allow_html=True)
-    
     names = []
-    st.markdown("""
-    <div class="container-fluid">
-        <div class="row align-items-start" style="padding-bottom:10px;">
-             <div  class="col-md-8 col-sm-8">
-                    <span>Company</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>Country</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>Customer Problem</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    <span>Business Model</span>
-             </div>
-             <div  class="col-md-1 col-sm-1">
-                    Actions
-             </div>
-        </div>             
-    """, unsafe_allow_html=True)
+    # list_html = """
+    # <h2>Companies list</h2>
+    # <div class="container-fluid">
+    #     <div class="row align-items-start" style="padding-bottom:10px;">
+    #          <div  class="col-md-8 col-sm-8">
+    #                 <span>Company</span>
+    #          </div>
+    #          <div  class="col-md-1 col-sm-1">
+    #                 <span>Country</span>
+    #          </div>
+    #          <div  class="col-md-1 col-sm-1">
+    #                 <span>Customer Problem</span>
+    #          </div>
+    #          <div  class="col-md-1 col-sm-1">
+    #                 <span>Business Model</span>
+    #          </div>
+    #          <div  class="col-md-1 col-sm-1">
+    #                 Actions
+    #          </div>
+    #     </div>             
+    # """
+    list_html = "<div class='container-fluid'>"
+
     for r in sorted_results:
         company_name = r["name"]
         if company_name in names:
@@ -235,41 +221,45 @@ def run_query(query, prompt, scrape_boost, top_k , regions, countries, is_debug,
         region = r["metadata"]["region"]
         country = r["metadata"]["country"]
         company_id = r["metadata"]["company_id"]
-        card(company_id, company_name, description, score, data_type, region, country, r['data'], is_debug)
+        list_html = list_html + card(company_id, company_name, description, score, data_type, region, country, r['data'], is_debug)
 
-    st.markdown('</div>')
+    list_html = list_html + '</div>'
+    st.markdown(list_html, unsafe_allow_html=True)
 
 
-def check_password():
-    """Returns `True` if the user had the correct password."""
+def render_history():
+    with st.session_state.history_container:
+
+        s = f"""
+            <div style='overflow: hidden;'>
+                <div id="chat_history" style='overflow-y: scroll;height: 100px;'>
+        """
+        for m in st.session_state.messages:
+            #print(f"Printing message\t {m['role']}: {m['content']}")        
+            s = s + f"<div>{m['role']}: {m['content']}</div>"
+        
+        s = s + f"""</div>
+            </div>
+            <script>
+                var el = document.getElementById("chat_history");
+                console.log(el.scrollTop, el.scrollHeight);
+                el.scrollTop = el.scrollHeight;
+                console.log(el.scrollTop, el.scrollHeight);
+            </script>            
+        """
+
+        components.html(s, height=140)
+        #st.markdown(s, unsafe_allow_html=True)
 
-    def password_entered():
-        """Checks whether a password entered by the user is correct."""
-        if st.session_state["password"] == st.secrets["password"]:
-            st.session_state["password_correct"] = True
-            del st.session_state["password"]  # don't store password
-        else:
-            st.session_state["password_correct"] = False
-
-    if "password_correct" not in st.session_state:
-        # First run, show input for password.
-        st.text_input(
-            "Password", type="password", on_change=password_entered, key="password"
-        )
-        return False
-    elif not st.session_state["password_correct"]:
-        # Password not correct, show input + error.
-        st.text_input(
-            "Password", type="password", on_change=password_entered, key="password"
-        )
-        st.error("😕 Password incorrect")
-        return False
-    else:
-        # Password correct.
-        return True
 
-if check_password():
-    st.title("")
+if utils.check_password():
+
+    st.markdown("<script language='javascript'>console.log('scrolling');</script>", unsafe_allow_html=True)
+
+    if "messages" not in st.session_state:
+        st.session_state.messages = [{"role":"system", "content":"You are an assistant who helps users find startups to invest in."}]        
+    
+    st.title("Raized")
 
     st.write("""
     Search for a company in free text. Describe the type of company you are looking for, the problem they solve and the solution they provide. You can also copy in the description of a similar company to kick off the search.
@@ -317,6 +307,8 @@ if check_password():
         ''',
         unsafe_allow_html=True
     )
+    st.session_state.history_container = st.container()
+
     tab_search, tab_advanced = st.tabs(["Search", "Settings"])
 
     
@@ -331,6 +323,7 @@ if check_password():
         scrape_boost = st.number_input('Web to API content ratio', value=1.)
         top_k = st.number_input('# Top Results', value=20)
         is_debug = st.checkbox("Debug output", value = False, key="debug")
+        openai_model = st.selectbox(label="Model", options=["gpt-4-1106-preview", "gpt-3.5-turbo-16k-0613", "gpt-3.5-turbo-16k"], index=0, key="openai_model")
         index_namespace = st.selectbox(label="Data Type", options=["websummarized", "web", "cbli", "all"], index=0)
         liked_companies = st.text_input(label="liked companies", key='liked_companies')
         disliked_companies = st.text_input(label="disliked companies", key='disliked_companies')
@@ -340,10 +333,16 @@ if check_password():
     with tab_search:    
         #report_type = st.multiselect("Report Type", utils.get_prompts(), key="search_prompts_multiselect")
         query = st.text_input("Search!", "")
-        cluster = st.checkbox("Cluster the results", value = False, key = "cluster")
+        #cluster = st.checkbox("Cluster the results", value = False, key = "cluster")
+        report_type = st.selectbox(label="Response Type", options=["company_list", "standard", "clustered"], index=0)
         #prompt_new = st.button("New", on_click = _prompt(prompt_title, prompt))        
 
     if query != "":
-        prompt = clustering_prompt if cluster else default_prompt
-        run_query(query, prompt, scrape_boost, top_k, region_selectbox, countries_selectbox, is_debug, index_namespace)
+        if report_type=="standard":
+            prompt = default_prompt
+        elif report_type=="clustered":
+            prompt = clustering_prompt 
+        else:
+            prompt = ""
+        run_query(query, prompt, scrape_boost, top_k, region_selectbox, countries_selectbox, is_debug, index_namespace, openai_model)
 

utils.py

@@ -2,6 +2,7 @@ import pandas as pd
 import psycopg2
 from psycopg2 import extras
 import streamlit as st
+import openai
 
 # def create_connection():
 #     host = st.secrets["RAIZED_DB_HOST"]
@@ -17,7 +18,51 @@ import streamlit as st
 #         )    
 
 ###
-        
+     
+def call_openai(prompt, engine="gpt-3.5-turbo", temp=0, top_p=1.0, max_tokens=4048):
+    try:
+        response = openai.ChatCompletion.create(
+            model=engine,
+            messages=st.session_state.messages,
+            temperature=temp,
+            max_tokens=max_tokens
+        )
+        print(f"Open AI response\n {response}")
+        text = response.choices[0].message["content"].strip()
+        st.session_state.messages.append({"role": "system", "content": text})
+        return text
+    except openai.error.OpenAIError as e:
+        print(f"An error occurred: {str(e)}")
+    return "Failed to generate a response."
+
+   
+def check_password():
+    """Returns `True` if the user had the correct password."""
+
+    def password_entered():
+        """Checks whether a password entered by the user is correct."""
+        if st.session_state["password"] == st.secrets["password"]:
+            st.session_state["password_correct"] = True
+            del st.session_state["password"]  # don't store password
+        else:
+            st.session_state["password_correct"] = False
+
+    if "password_correct" not in st.session_state:
+        # First run, show input for password.
+        st.text_input(
+            "Password", type="password", on_change=password_entered, key="password"
+        )
+        return False
+    elif not st.session_state["password_correct"]:
+        # Password not correct, show input + error.
+        st.text_input(
+            "Password", type="password", on_change=password_entered, key="password"
+        )
+        st.error("😕 Password incorrect")
+        return False
+    else:
+        # Password correct.
+        return True
 
 def get_prompt(title):
     return ""
@@ -35,14 +80,21 @@ def get_prompt(title):
     # print(f"Results getting {title}")
     # return res
 
+# default_prompt = """
+# summarize the outcome of this search. The context is a list of company names followed by the company's description and a relevance score to the user query. 
+# the report should mention the most important companies and how they compare to each other and contain the following sections:
+# 1) Title: query text (summarized if more than 20 tokens)
+# 2) Best matches: Naming of the 3 companies from the list that are most similar to the search query:
+# - summarize what they are doing
+# - name customers and technology if they are mentioned
+# - compare them to each other and point out what they do differently or what is their unique selling proposition
+# ----"""    
 default_prompt = """
-summarize the outcome of this search. The context is a list of company names followed by the company's description and a relevance score to the user query. 
-the report should mention the most important companies and how they compare to each other and contain the following sections:
-1) Title: query text (summarized if more than 20 tokens)
-2) Best matches: Naming of the 3 companies from the list that are most similar to the search query:
-- summarize what they are doing
+You are an invesment assistant. Below is a user query followed by a list of company descriptions that match the user query.
+the report should mention the most important companies and how they compare to each other and contain the following sections
+- summarize what those companies they are doing
 - name customers and technology if they are mentioned
-- compare them to each other and point out what they do differently or what is their unique selling proposition
+- compare the companies to each other and point out what they do differently or what is their unique selling proposition
 ----"""    
 
 clustering_prompt = """Please create a document with the following headings: 
@@ -76,4 +128,21 @@ List with all the companies in this cluster. Each list item should be structured
 * name of the company in bold (URL of the company, country location of the company): short summary summary of what the company does (max 30 tokens)
 H1: How you could improve your search
 “I hope you have already found some interesting matches. I am happy to let you refine your search. Here are some ideas on how to find matches in relation to your original question around (“user query”):”
-* List of ideas on how to refine and improve the search"""
+* List of ideas on how to refine and improve the search"""
+
+
+
+
+def on_prompt_selected():
+    title = st.session_state.advanced_prompts_select
+    new_prompt = utils.get_prompt(title)
+    if len(new_prompt)>0 and len(new_prompt[0])>0:
+        print(f"Got a prompt for title {title}\n {new_prompt[0]}")
+        st.session_state.prompt_title_editable = st.session_state.advanced_prompts_select
+        st.session_state.advanced_prompt_content = new_prompt[0]
+    else:
+        print(f"No results for title {st.session_state.advanced_prompts_select}")
+
+
+
+