Update app.py

app.py

@@ -1,144 +1,242 @@
 import openai
-import numpy as np
-import pandas as pd
-import os
-from langchain.embeddings.openai import OpenAIEmbeddings
-from langchain.embeddings.huggingface import HuggingFaceEmbeddings
-from langchain import HuggingFaceHub
-from langchain.vectorstores import Chroma
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain.llms import OpenAI
-from langchain.chains import RetrievalQA
-from langchain.chains import VectorDBQA
-from langchain.document_loaders import TextLoader, WebBaseLoader, SeleniumURLLoader
-from langchain.document_loaders import UnstructuredFileLoader
+
+openai.api_key=os.getenv("OPENAI_API_KEY")
+
+from dotenv import load_dotenv
+load_dotenv()
+
 from flask import Flask, jsonify, render_template, request
+import requests, json
+
+# import nltk
+# nltk.download("punkt")
+
+import os
+import shutil
 from werkzeug.utils import secure_filename
-from werkzeug.datastructures import  FileStorage
+from werkzeug.datastructures import FileStorage
+
 import nltk
-nltk.download("punkt")
-import warnings
-warnings.filterwarnings("ignore")
 
+from datetime import datetime
 
+import openai
+from langchain.llms import OpenAI
 
-openai.api_key=os.getenv("OPENAI_API_KEY")
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
 
+from langchain.document_loaders import SeleniumURLLoader, PyPDFLoader
 
-import flask
-import os
-from dotenv import load_dotenv
-load_dotenv()
+from langchain.vectorstores import Chroma
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+
+from langchain.chains import VectorDBQA
+
+from langchain.document_loaders import UnstructuredFileLoader
+from langchain import PromptTemplate
+
+from langchain.chains import RetrievalQA
+from langchain.memory import ConversationBufferWindowMemory
 
-loader = UnstructuredFileLoader('Jio.txt', mode='elements')
-documents= loader.load()
+import warnings
+
+warnings.filterwarnings("ignore")
 
-text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=150)
-texts = text_splitter.split_documents(documents)
-embeddings = OpenAIEmbeddings()
-vectordb = Chroma.from_documents(texts,embeddings)
-chain = RetrievalQA.from_chain_type(llm=OpenAI(temperature=0.0),chain_type="stuff", retriever=vectordb.as_retriever(search_type="mmr"),return_source_documents=True)
 
+#app = Flask(__name__)
 
 app = flask.Flask(__name__, template_folder="./")
+
 # Create a directory in a known location to save files to.
 uploads_dir = os.path.join(app.root_path,'static', 'uploads')
+
 os.makedirs(uploads_dir, exist_ok=True)
 
+vectordb = createVectorDB(loadKB(False, False, uploads_dir, None))
+
+@app.route('/', methods=['GET'])
+def test():
+    return "Docker hello"
+
+@app.route('/KBUploader')
+def KBUpload():
+    return render_template("FileUpload.html")
 
-@app.route('/Home')
-def index():
-    return flask.render_template('index.html')
+@app.route('/aiassist')
+def aiassist():
+    return render_template("AIAssist.html")
 
-@app.route('/post_json', methods=['POST'])
+@app.route('/agent/chat/suggestion', methods=['POST'])
 def process_json():
+    print(f"\n{'*' * 100}\n")
+    print("Request Received >>>>>>>>>>>>>>>>>>", datetime.now().strftime("%H:%M:%S"))
     content_type = request.headers.get('Content-Type')
     if (content_type == 'application/json'):
-        
-        userQuery = request.get_json()['query']
-        responseJSON=chain({"query":userQuery});
-        print("Retrieved Document List START ***********************\n\n")
-        pretty_print_docs(responseJSON['source_documents'])
-        print("Retrieved Document List END ***********************\n\n")
-        print("Ques:>>>>"+userQuery+"\n Ans:>>>"+responseJSON["result"])
-        return jsonify(botMessage=responseJSON["result"]); 
+        requestQuery = request.get_json()
+        print(type(requestQuery))
+        custDetailsPresent=False
+        customerName=""
+        customerDistrict=""
+        if("custDetails" in requestQuery):
+            custDetailsPresent = True
+            customerName=requestQuery['custDetails']['cName']
+            customerDistrict=requestQuery['custDetails']['cDistrict']
+
+        print("chain initiation")
+        chainRAG=getRAGChain(customerName, customerDistrict, custDetailsPresent,vectordb)
+        print("chain created")
+        suggestionArray = []
+
+        for index, query in enumerate(requestQuery['message']):
+            #message = answering(query)
+            relevantDoc = vectordb.similarity_search_with_score(query)
+            for doc in relevantDoc:
+                print(f"\n{'-' * 100}\n")
+                print("Document Source>>>>>>  " + doc[len(doc) - 2].metadata['source'] + "\n\n")
+                print("Page Content>>>>>> " + doc[len(doc) - 2].page_content + "\n\n")
+                print("Similarity Score>>>> " + str(doc[len(doc) - 1]))
+                print(f"\n{'-' * 100}\n")
+            message = chainRAG.run({"query": query})
+            print("query:",query)
+            print("Response:", message)
+            if "I don't know" in message:
+                 message = "Dear Sir/ Ma'am, Could you please ask questions relevant to Jio?"
+            responseJSON={"message":message,"id":index}
+            suggestionArray.append(responseJSON)
+        return jsonify(suggestions=suggestionArray)
     else:
         return 'Content-Type not supported!'
 
-
-@app.route('/file_upload',methods=['POST'])
+@app.route('/file_upload', methods=['POST'])
 def file_Upload():
-        
-        fileprovided=not request.files.getlist('files[]')[0].filename==''
-        urlProvided=not request.form.getlist('weburl')[0]==''
-        print("*******")
-        print("File Provided:"+str(fileprovided))
-        print("URL Provided:"+str(urlProvided))
-        print("*******")
-        print(not ('documents' in vars() or 'documents' in globals()))
-        # if not ('documents' in vars() or 'documents' in globals()):
-        documents = []
-        if fileprovided:
-            
-            #Delete Files
-            for filename in os.listdir(uploads_dir):
-                file_path = os.path.join(uploads_dir, filename)
-                print("Clearing Doc Directory. Trying to delete"+file_path)
-                try:
-                    if os.path.isfile(file_path) or os.path.islink(file_path):
-                        os.unlink(file_path)
-                    elif os.path.isdir(file_path):
-                        shutil.rmtree(file_path)
-                except Exception as e:
-                    print('Failed to delete %s. Reason: %s' % (file_path, e))
-            #Read and Embed New Files provided        
-            for file in request.files.getlist('files[]'):
-                print(file.filename)
-                file.save(os.path.join(uploads_dir, secure_filename(file.filename)))
-                loader = UnstructuredFileLoader(os.path.join(uploads_dir, secure_filename(file.filename)), mode='elements')
-                documents.extend(loader.load())
-        else:
-             loader = UnstructuredFileLoader('Jio.txt', mode='elements')
-             documents.extend(loader.load())
-
-        if urlProvided:
-            weburl=request.form.getlist('weburl')
-            print(weburl)
-            urlList=weburl[0].split(';')
-            print(urlList)
-            urlLoader=SeleniumURLLoader(urlList)
-            documents.extend(urlLoader.load())
-
-
-    
-        print(uploads_dir)
-        global chain;
-
-        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=150)
-        texts = text_splitter.split_documents(documents)
-        
-        print("All chunk List START ***********************\n\n")
-        pretty_print_docs(texts)
-        print("All chunk List END ***********************\n\n")
-        
-        embeddings = OpenAIEmbeddings()
-        vectordb = Chroma.from_documents(texts,embeddings)
-        chain = RetrievalQA.from_chain_type(llm=OpenAI(temperature=0.0),chain_type="stuff", retriever=vectordb.as_retriever(search_type="mmr"),return_source_documents=True)
-        
-        return render_template("index.html")
-
-@app.route('/')
-def KBUpload():
-    return render_template("KBTrain.html")
+    fileprovided = not request.files.getlist('files[]')[0].filename == ''
+    urlProvided = not request.form.getlist('weburl')[0] == ''
+    print("*******")
+    print("File Provided:" + str(fileprovided))
+    print("URL Provided:" + str(urlProvided))
+    print("*******")
+
+    print(uploads_dir)
+    documents = loadKB(fileprovided, urlProvided, uploads_dir, request)
+    vectordb=createVectorDB(documents)
+    return render_template("AIAssist.html")
+
+def createPrompt(cName, cCity, custDetailsPresent):
+    cProfile = "Customer's Name is " + cName + "\nCustomer's lives in or customer's Resident State or Customer's place is " + cCity + "\n"
+    print(cProfile)
+
+    template1 = """You role is of a Professional Customer Support Executive and your name is Jio AIAssist.
+        You are talking to the below customer whose information is provided in block delimited by <cp></cp>.
+        Use the following customer related information (delimited by <cp></cp>) and context (delimited by <ctx></ctx>) to answer the question at the end by thinking step by step alongwith reaonsing steps:
+        If you don't know the answer, just say that you don't know, don't try to make up an answer.
+        Use the customer information to replace entities in the question before answering\n
+        \n"""
+
+    template2 = """
+        <ctx>
+        {context}
+        </ctx>
+        <hs>
+        {history}
+        </hs>
+        Question: {question}
+        Answer: """
+
+    prompt_template = template1 + "<cp>\n" + cProfile + "\n</cp>\n" + template2
+    PROMPT = PromptTemplate(template=prompt_template, input_variables=["history", "context", "question"])
+    return PROMPT
 
-@app.route('/aiassist')
-def aiassist():
-    return render_template("index.html")
 
 def pretty_print_docs(docs):
-    print(f"\n{'-' * 100}\n".join([f"Document {i+1}:\n\n" + "Document Source>>> "+d.metadata['source']+"\n\nContent>>> "+d.page_content for i, d in enumerate(docs)]))
+    print(f"\n{'-' * 100}\n".join([f"Document {i + 1}:\n\n" + "Document Length>>>" + str(
+        len(d.page_content)) + "\n\nDocument Source>>> " + d.metadata['source'] + "\n\nContent>>> " + d.page_content for
+                                   i, d in enumerate(docs)]))
 
 
+def getEmbeddingModel(embeddingId):
+    if (embeddingId == 1):
+        embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
+    else:
+        embeddings = OpenAIEmbeddings()
+    return embeddings
+
+
+def getLLMModel(LLMID):
+    llm = OpenAI(temperature=0.0)
+    return llm
+
+
+def clearKBUploadDirectory(uploads_dir):
+    for filename in os.listdir(uploads_dir):
+        file_path = os.path.join(uploads_dir, filename)
+        print("Clearing Doc Directory. Trying to delete" + file_path)
+        try:
+            if os.path.isfile(file_path) or os.path.islink(file_path):
+                os.unlink(file_path)
+            elif os.path.isdir(file_path):
+                shutil.rmtree(file_path)
+        except Exception as e:
+            print('Failed to delete %s. Reason: %s' % (file_path, e))
+
+
+def loadKB(fileprovided, urlProvided, uploads_dir, request):
+    documents = []
+    if fileprovided:
+        # Delete Files
+        clearKBUploadDirectory(uploads_dir)
+        # Read and Embed New Files provided
+        for file in request.files.getlist('files[]'):
+            print("File Received>>>" + file.filename)
+            file.save(os.path.join(uploads_dir, secure_filename(file.filename)))
+            loader = PyPDFLoader(os.path.join(uploads_dir, secure_filename(file.filename)))
+            documents.extend(loader.load())
+    else:
+        loader = PyPDFLoader('./KnowledgeBase/Jio.pdf')
+        documents.extend(loader.load())
+
+    if urlProvided:
+        weburl = request.form.getlist('weburl')
+        print(weburl)
+        urlList = weburl[0].split(';')
+        print(urlList)
+        print("Selenium Started", datetime.now().strftime("%H:%M:%S"))
+        # urlLoader=RecursiveUrlLoader(urlList[0])
+        urlLoader = SeleniumURLLoader(urlList)
+        print("Selenium Completed", datetime.now().strftime("%H:%M:%S"))
+        documents.extend(urlLoader.load())
+
+    return documents
+
+
+def getRAGChain(customerName,customerDistrict, custDetailsPresent,vectordb):
+    chain = RetrievalQA.from_chain_type(
+        llm=getLLMModel(0),
+        chain_type='stuff',
+        retriever=vectordb.as_retriever(),
+        verbose=False,
+        chain_type_kwargs={
+            "verbose": False,
+            "prompt": createPrompt(customerName, customerDistrict, custDetailsPresent),
+            "memory": ConversationBufferWindowMemory(
+                k=3,
+                memory_key="history",
+                input_key="question"),
+        }
+    )
+    return chain
+
+def createVectorDB(documents):
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=150)
+    texts = text_splitter.split_documents(documents)
+    print("All chunk List START ***********************\n\n")
+    pretty_print_docs(texts)
+    print("All chunk List END ***********************\n\n")
+    embeddings = getEmbeddingModel(0)
+    vectordb = Chroma.from_documents(texts, embeddings)
+    return vectordb
+
 
 if __name__ == '__main__':
     app.run(host='0.0.0.0',  port=int(os.environ.get('PORT', 7860)))
+