agent implementation

app.py

@@ -132,11 +132,16 @@ def parameters_change_button(chat_model, show_source):
 
 
 def get_answer_from_backend(query, model, dataset):
-    response = qaPipeline.run(query=query, model=model, dataset=dataset)
+    # response = qaPipeline.run(query=query, model=model, dataset=dataset)
+    response = qaPipeline.run_agent(query=query, model=model, dataset=dataset)
     return response
 
 def show_query_response(query, response, show_source_files):
-    answer, docs = response['result'], response['source_documents']
+    docs = []
+    if isinstance(response, dict):
+        answer, docs = response['result'], response['source_documents']
+    else:
+        answer = response
  
     st.write(user_template.replace(
                 "{{MSG}}", query), unsafe_allow_html=True)

qaPipeline.py

@@ -23,6 +23,11 @@ from langchain.chat_models import ChatOpenAI
 # from chromaDb import load_store
 from faissDb import load_FAISS_store
 
+from langchain.agents import initialize_agent, Tool
+from langchain.agents import AgentType
+from langchain.prompts import PromptTemplate
+from langchain.chains import LLMChain
+
 load_dotenv()
 
 #gpt4 all model
@@ -49,6 +54,7 @@ class QAPipeline:
         self.vectorstore = None
 
         self.qa_chain = None
+        self.agent = None
 
     def run(self,query, model, dataset):
 
@@ -70,6 +76,28 @@ class QAPipeline:
         print( res)
 
         return res
+    
+    def run_agent(self,query, model, dataset):
+
+        if (self.llm_name != model) or (self.dataset_name != dataset) or (self.agent == None):
+            self.set_model(model)
+            self.set_vectorstore(dataset)
+            self.set_qa_chain_with_agent()
+
+        # Get the answer from the chain
+        start = time.time()
+        res = self.agent(query)
+        # answer, docs = res['result'],res['source_documents']
+        end = time.time()
+
+        # Print the result
+        print("\n\n> Question:")
+        print(query)
+        print(f"\n> Answer (took {round(end - start, 2)} s.):")
+        print( res)
+
+        return res["output"]
+
 
     
     def set_model(self,model_type):
@@ -107,4 +135,71 @@ class QAPipeline:
             # retriever = self.vectorstore.as_retriever(search_kwargs={"k": target_source_chunks}
             return_source_documents= True
         )
-        
+
+
+    def set_qa_chain_with_agent(self):
+
+        # Define a custom prompt
+        general_qa_template = (
+            """You are the AI assistant of the Boardpac company which provide services for companies board members.
+            You can have a general conversation with the users like greetings.
+            But only answer questions related to banking sector like financial and legal.
+            If you dont know the answer say you dont know, dont try to makeup answers.
+            each answer should start with code word BoardPac Conversation AI:
+            Question: {question}
+            """
+        )
+
+        general_qa_chain_prompt = PromptTemplate.from_template(general_qa_template)
+        general_qa_chain = LLMChain(llm=self.llm, prompt=general_qa_chain_prompt)
+
+        # Define a custom prompt
+        retrieval_qa_template = (
+            """You are the AI assistant of the Boardpac company which provide services for companies board members.
+            You have provided context information below related to central bank acts published in various years. The content of a bank act can updated by a bank act from a latest year.
+            {context}
+            Given this information, please answer the question with the latest information. 
+            If you dont know the answer say you dont know, dont try to makeup answers. 
+            each answer should start with code word BoardPac Retrieval AI:
+            Question: {question}
+            """
+        )
+        retrieval_qa_chain_prompt = PromptTemplate.from_template(retrieval_qa_template)
+
+        bank_regulations_qa = RetrievalQA.from_chain_type(
+            llm=self.llm,
+            chain_type="stuff",
+            retriever = self.vectorstore.as_retriever(), 
+            # retriever = self.vectorstore.as_retriever(search_kwargs={"k": target_source_chunks}
+            return_source_documents= True,
+            input_key="question",
+            chain_type_kwargs={"prompt": retrieval_qa_chain_prompt},
+        )
+
+        tools = [
+            Tool(
+                name="bank regulations",
+                func= lambda query: bank_regulations_qa({"question": query}),
+                description='''useful for when you need to answer questions about 
+                financial and legal information issued from central bank regarding banks and bank regulations. 
+                Input should be a fully formed question.''',
+                return_direct=True,
+            ),
+
+            Tool(
+                name="general qa",
+                func= general_qa_chain.run,
+                description='''useful for when you need to have a general conversation with the users like greetings 
+                or to answer general purpose questions related to banking sector like financial and legal. 
+                Input should be a fully formed question.''',
+                return_direct=True,
+            ),  
+        ]
+
+        self.agent = initialize_agent(
+            tools, 
+            self.llm, 
+            agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, 
+            verbose=True,
+            max_iterations=3,
+        )