Merge pull request #9 from robertmcleod2/add-rag

Add rag

.github/workflows/deploy.yml

@@ -20,12 +20,13 @@ jobs:
       contents: read
     # retrieve variables from environment secrets and variables
     env:
-      # deplyoment variables
+      # deployment variables
       AZURE_SUBSCRIPTION_ID: ${{ vars.AZURE_SUBSCRIPTION_ID }}
       AZURE_TENANT_ID: ${{ vars.AZURE_TENANT_ID }}
       AZURE_CLIENT_ID: ${{ vars.AZURE_CLIENT_ID }}
-      AZURE_CLIENT_CERT_NAME: ${{ vars.AZURE_CLIENT_CERT_NAME }}
-      AZURE_CLIENT_CERT: ${{ secrets.AZURE_CLIENT_CERT }}
+      # blank variable to store the certificate content
+      AZURE_CLIENT_CERTIFICATE_PATH: "blank"
+      AZURE_CLIENT_CERTIFICATE: ${{ secrets.AZURE_CLIENT_CERTIFICATE }}
       AZURE_RESOURCE_GROUP: ${{ vars.AZURE_RESOURCE_GROUP }}  
       # App env variables
       OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
@@ -38,11 +39,11 @@ jobs:
       - name: run deploy.sh
         shell: bash
         run: |
-          echo "$AZURE_CLIENT_CERT" > ${{ vars.AZURE_CLIENT_CERT_NAME }}
+          echo "$AZURE_CLIENT_CERTIFICATE" > ${AZURE_CLIENT_CERTIFICATE_PATH}
           bash deployment/deploy.sh
       - name: clean up
         if: always()
         shell: bash
         run: |
           az logout
-          rm ${{ vars.AZURE_CLIENT_CERT_NAME }}
+          rm ${AZURE_CLIENT_CERTIFICATE_PATH}

README.md

@@ -34,4 +34,6 @@ pip install -r requirements_dev.txt
 
 ```bash
 streamlit run src/app.py
-```
+```
+
+6. View the application in your browser at `http://localhost:8501`. The password is the one you set in the `.env` file.

deployment/deploy.sh

@@ -4,7 +4,7 @@ export $(egrep -v '^#' .env | xargs)
 LOCATION=westeurope
 
 # log in to Azure with service principal
-az login --service-principal --username ${AZURE_CLIENT_ID} --password ${AZURE_CLIENT_CERT_NAME} --tenant ${AZURE_TENANT_ID}
+az login --service-principal --username ${AZURE_CLIENT_ID} --password ${AZURE_CLIENT_CERTIFICATE_PATH} --tenant ${AZURE_TENANT_ID}
 az account set --subscription ${AZURE_SUBSCRIPTION_ID}
 az config set defaults.group=${AZURE_RESOURCE_GROUP} defaults.location=${LOCATION}
 

deployment/setup_env.sh

@@ -4,7 +4,7 @@ export $(egrep -v '^#' .env | xargs)
 LOCATION=westeurope
 
 # log in to Azure with service principal
-az login --service-principal --username ${AZURE_CLIENT_ID} --password ${AZURE_CLIENT_CERT_NAME} --tenant ${AZURE_TENANT_ID}
+az login --service-principal --username ${AZURE_CLIENT_ID} --password ${AZURE_CLIENT_CERTIFICATE_PATH} --tenant ${AZURE_TENANT_ID}
 az account set --subscription ${AZURE_SUBSCRIPTION_ID}
 az config set defaults.group=${AZURE_RESOURCE_GROUP} defaults.location=${LOCATION}
 

basic_chatbot.ipynb → notebooks/basic_chatbot.ipynb

@@ -43,11 +43,13 @@
     "\n",
     "store = {}\n",
     "\n",
+    "\n",
     "def get_session_history(session_id: str) -> BaseChatMessageHistory:\n",
     "    if session_id not in store:\n",
     "        store[session_id] = InMemoryChatMessageHistory()\n",
     "    return store[session_id]\n",
     "\n",
+    "\n",
     "trimmer = trim_messages(\n",
     "    max_tokens=1000,\n",
     "    strategy=\"last\",\n",
@@ -66,15 +68,13 @@
     "            The user will describe their energy usage and you will help them to detect anomalies.\\n\n",
     "            You will also help the user to identify the causes of the anomalies \\n\n",
     "            and suggest ways to fix them.\\n\n",
-    "            \"\"\"\n",
+    "            \"\"\",\n",
     "        ),\n",
     "        MessagesPlaceholder(variable_name=\"messages\"),\n",
     "    ]\n",
     ")\n",
     "\n",
-    "chain = (\n",
-    "    trimmer | prompt | model\n",
-    ")\n",
+    "chain = trimmer | prompt | model\n",
     "\n",
     "with_message_history = RunnableWithMessageHistory(\n",
     "    chain,\n",

exploration.ipynb → notebooks/exploration.ipynb

@@ -106,15 +106,19 @@
    "outputs": [],
    "source": [
     "response = with_message_history.invoke(\n",
-    "    [HumanMessage(content=\"\"\"list the names of people who have said hello to you in our conversation so far.\\n\n",
+    "    [\n",
+    "        HumanMessage(\n",
+    "            content=\"\"\"list the names of people who have said hello to you in our conversation so far.\\n\n",
     "List the names following the below format:\\n\n",
     "\n",
     "[<Name 1>, <Name 2>, <Name 3>, ...]\n",
     "\n",
     "For example, if the names of people who have said hello to you in our conversation so far are Alice and Bob, you should list them as follows:\n",
     "    \n",
     "[Alice, Bob]                  \n",
-    "\"\"\")],\n",
+    "\"\"\"\n",
+    "        )\n",
+    "    ],\n",
     "    config=config,\n",
     ")\n",
     "\n",
@@ -291,11 +295,7 @@
     "    AIMessage(content=\"yes!\"),\n",
     "]\n",
     "\n",
-    "chain = (\n",
-    "    RunnablePassthrough.assign(messages=itemgetter(\"messages\") | trimmer)\n",
-    "    | prompt\n",
-    "    | model\n",
-    ")\n",
+    "chain = RunnablePassthrough.assign(messages=itemgetter(\"messages\") | trimmer) | prompt | model\n",
     "\n",
     "with_message_history = RunnableWithMessageHistory(\n",
     "    chain,\n",
@@ -313,7 +313,7 @@
     "    config=config,\n",
     ")\n",
     "\n",
-    "response.content\n"
+    "response.content"
    ]
   },
   {
@@ -339,6 +339,200 @@
     "):\n",
     "    print(r.content, end=\"\")"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Chatbot with memory and RAG\n",
+    "\n",
+    "chatbot uses retriever to get relevant information to query from vector store, for every query. \n",
+    "\n",
+    "this means we are taking a while to get the information, but we are getting the most relevant information. \n",
+    "\n",
+    "probably not the most useful for our application as for now we are normally just going to want to get the information related to energy anomaly detection and use that every time as context.\n",
+    "\n",
+    "Maybe we would want to give the chatbot agent the ability to decide if it wants to retrive information from the vector store or not."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 41,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "\n",
+    "from dotenv import load_dotenv\n",
+    "from langchain.chains import create_history_aware_retriever, create_retrieval_chain\n",
+    "from langchain.chains.combine_documents import create_stuff_documents_chain\n",
+    "from langchain.globals import set_debug\n",
+    "from langchain_community.chat_message_histories import ChatMessageHistory\n",
+    "from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder\n",
+    "from langchain_core.runnables.history import RunnableWithMessageHistory\n",
+    "from langchain_core.vectorstores import InMemoryVectorStore\n",
+    "from langchain_openai import ChatOpenAI, OpenAIEmbeddings\n",
+    "from langchain_text_splitters import RecursiveJsonSplitter\n",
+    "\n",
+    "llm = ChatOpenAI(model=\"gpt-4o-mini\", temperature=0)\n",
+    "\n",
+    "### Construct retriever ###\n",
+    "with open(\"../src/example_customer_documents.json\") as f:\n",
+    "    json_data = json.load(f)\n",
+    "\n",
+    "splitter = RecursiveJsonSplitter(max_chunk_size=300)\n",
+    "docs = splitter.create_documents(texts=[json_data])\n",
+    "\n",
+    "vectorstore = InMemoryVectorStore.from_documents(documents=docs, embedding=OpenAIEmbeddings())\n",
+    "retriever = vectorstore.as_retriever()\n",
+    "\n",
+    "\n",
+    "### Contextualize question ###\n",
+    "contextualize_q_system_prompt = \"\"\"Given a chat history and the latest user question \\\n",
+    "which might reference context in the chat history, formulate a standalone question \\\n",
+    "which can be understood without the chat history. \\\n",
+    "Include information relevant to energy anomaly detection in the question, if needed. \\\n",
+    "Do NOT answer the question, just reformulate it if needed and otherwise return it as is.\"\"\"\n",
+    "contextualize_q_prompt = ChatPromptTemplate.from_messages(\n",
+    "    [\n",
+    "        (\"system\", contextualize_q_system_prompt),\n",
+    "        MessagesPlaceholder(\"chat_history\"),\n",
+    "        (\"human\", \"{input}\"),\n",
+    "    ]\n",
+    ")\n",
+    "history_aware_retriever = create_history_aware_retriever(llm, retriever, contextualize_q_prompt)\n",
+    "\n",
+    "\n",
+    "### Answer question ###\n",
+    "qa_system_prompt = \"\"\"You are an an energy usage anomaly detection assistant. \\\n",
+    "You are helping a user to detect anomalies in their energy usage. \\\n",
+    "The user will describe their energy usage and you will help them to detect anomalies. \\\n",
+    "You will also help the user to identify the causes of the anomalies \\\n",
+    "and suggest ways to fix them. \\\n",
+    "\n",
+    "Use the following pieces of retrieved context to answer the question if needed. \\\n",
+    "Follow up on previous parts of customer service chatbot and agent conversations that are not yet resolved. \\\n",
+    "\n",
+    "{context}\"\"\"\n",
+    "\n",
+    "qa_prompt = ChatPromptTemplate.from_messages(\n",
+    "    [\n",
+    "        (\"system\", qa_system_prompt),\n",
+    "        MessagesPlaceholder(\"chat_history\"),\n",
+    "        (\"human\", \"{input}\"),\n",
+    "    ]\n",
+    ")\n",
+    "question_answer_chain = create_stuff_documents_chain(llm, qa_prompt)\n",
+    "\n",
+    "rag_chain = create_retrieval_chain(history_aware_retriever, question_answer_chain)\n",
+    "\n",
+    "\n",
+    "### Statefully manage chat history ###\n",
+    "chat_history = ChatMessageHistory()\n",
+    "\n",
+    "\n",
+    "conversational_rag_chain = RunnableWithMessageHistory(\n",
+    "    rag_chain,\n",
+    "    lambda session_id: chat_history,\n",
+    "    input_messages_key=\"input\",\n",
+    "    history_messages_key=\"chat_history\",\n",
+    "    output_messages_key=\"answer\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.globals import set_debug\n",
+    "\n",
+    "set_debug(False)\n",
+    "\n",
+    "conversational_rag_chain.invoke({\"input\": \"I increased my thermostat temperature during the day\"}, {\"configurable\": {\"session_id\": \"unused\"}})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Chatbot that is passed a fixed context RAG\n",
+    "\n",
+    "Simplest implementation of RAG: just passing a fixed context from the vector store using the retriever and a fixed query. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 51,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "\n",
+    "from dotenv import load_dotenv\n",
+    "from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder\n",
+    "from langchain_core.runnables.history import RunnableWithMessageHistory\n",
+    "from langchain_core.vectorstores import InMemoryVectorStore\n",
+    "from langchain_openai import ChatOpenAI, OpenAIEmbeddings\n",
+    "from langchain_text_splitters import RecursiveJsonSplitter\n",
+    "\n",
+    "load_dotenv()\n",
+    "\n",
+    "model = ChatOpenAI(model=\"gpt-4o-mini\")\n",
+    "\n",
+    "# local document retrieval\n",
+    "with open(\"../src/example_customer_documents.json\") as f:\n",
+    "    json_data = json.load(f)\n",
+    "\n",
+    "splitter = RecursiveJsonSplitter(max_chunk_size=300)\n",
+    "docs = splitter.create_documents(texts=[json_data])\n",
+    "\n",
+    "vectorstore = InMemoryVectorStore.from_documents(documents=docs, embedding=OpenAIEmbeddings())\n",
+    "retriever = vectorstore.as_retriever()\n",
+    "\n",
+    "context = retriever.invoke(\"energy anomaly detection smart meter energy consumption\")\n",
+    "\n",
+    "prompt = ChatPromptTemplate.from_messages(\n",
+    "    [\n",
+    "        (\n",
+    "            \"system\",\n",
+    "            \"\"\"You are an an energy usage anomaly detection assistant.\\n\n",
+    "            You are helping a user to detect anomalies in their energy usage.\\n\n",
+    "            The user will describe their energy usage and you will help them to detect anomalies.\\n\n",
+    "            You will also help the user to identify the causes of the anomalies \\n\n",
+    "            and suggest ways to fix them.\\n\n",
+    "\n",
+    "            Use the following pieces of retrieved context to answer the question if needed.\\n\n",
+    "            {context}\n",
+    "            \"\"\",\n",
+    "        ),\n",
+    "        MessagesPlaceholder(variable_name=\"chat_history\"),\n",
+    "        (\"human\", \"{input}\"),\n",
+    "    ]\n",
+    ")\n",
+    "\n",
+    "chain = prompt | model\n",
+    "\n",
+    "### Statefully manage chat history ###\n",
+    "chat_history = ChatMessageHistory()\n",
+    "\n",
+    "chain_with_message_history = RunnableWithMessageHistory(\n",
+    "    chain,\n",
+    "    lambda session_id: chat_history,\n",
+    "    input_messages_key=\"input\",\n",
+    "    history_messages_key=\"chat_history\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "chain_with_message_history.invoke({\"context\": context, \"input\": \"I increased my thermostat temperature during the day\"}, {\"configurable\": {\"session_id\": \"unused\"}})"
+   ]
   }
  ],
  "metadata": {

src/app.py

@@ -1,5 +1,5 @@
 import streamlit as st
-from basic_chatbot import setup_chain
+from chatbot_rag import ChatbotRAG as Chatbot
 from dotenv import load_dotenv
 from utils import check_password
 
@@ -10,8 +10,8 @@
 
 st.title("Energy Usage Anomaly Detection Assistant")
 
-if "chain" not in st.session_state:
-    st.session_state.chain = setup_chain()
+if "chatbot" not in st.session_state:
+    st.session_state.chatbot = Chatbot()
 
 if "messages" not in st.session_state:
     st.session_state.messages = []
@@ -26,6 +26,5 @@
         st.markdown(prompt)
 
     with st.chat_message("assistant"):
-        stream = st.session_state.chain.stream({"input": prompt}, {"configurable": {"session_id": "unused"}})
-        response = st.write_stream(stream)
+        response = st.session_state.chatbot.stream(prompt)
     st.session_state.messages.append({"role": "assistant", "content": response})