L3_Advanced_queries_with_Agents.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "90710518",
   "metadata": {},
   "source": [
    "# L3: Advanced queries with Agents"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ee47ffb3",
   "metadata": {},
   "source": [
    "In this lesson, you are going to do more advanced querying."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0850beba",
   "metadata": {},
   "source": [
    "### Import required dependencies and load the API key"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "0eb492b6-b561-4f95-882e-2d438692b21f",
   "metadata": {
    "height": 200
   },
   "outputs": [],
   "source": [
    "import * as mod from \"https://deno.land/std@0.213.0/dotenv/mod.ts\";\n",
    "import { \n",
    "    Document, \n",
    "    VectorStoreIndex, \n",
    "    SimpleDirectoryReader,\n",
    "    RouterQueryEngine,\n",
    "    OpenAIAgent,\n",
    "    QueryEngineTool,\n",
    "    FunctionTool,\n",
    "} from \"npm:llamaindex@0.1.18\"\n",
    "const keys = await mod.load({export:true}) // read API key from .env"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "57a8df97",
   "metadata": {},
   "source": [
    "Let's create 2 different data sources that answer different types of questions:\n",
    "- The first is the same example as in lesson 1, an essay by Dan Abramov about his career.\n",
    "- The second is the wikipedia page content about React.\n",
    "\n",
    "**Note**: To access the data used in this lesson, go to `File` and click on `Open` "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "838e4302-d735-48b6-bbec-a75dcec5cb15",
   "metadata": {
    "height": 64
   },
   "outputs": [],
   "source": [
    "const documents1 = await new SimpleDirectoryReader().loadData({directoryPath: \"./data\"})\n",
    "const index1 = await VectorStoreIndex.fromDocuments(documents1)\n",
    "const queryEngine1 = index1.asQueryEngine()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d4915eb5-98fe-4233-98cf-696d653a8552",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "In college, the author picked a college that gave Macbooks to students, but there wasn't much programming in the curriculum for two more years. The author started slacking off, skipping lectures, and eventually stopped attending classes altogether. The author returned the Macbook and only went back to the college five years later to pick up their papers.\n"
     ]
    }
   ],
   "source": [
    "let response1 = await queryEngine1.query({query: \"What did the author do in college?\"})\n",
    "console.log(response1.toString())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "d2906d8e-c949-45a0-a496-f6b65f7cf846",
   "metadata": {
    "height": 64
   },
   "outputs": [],
   "source": [
    "const documents2 = await new SimpleDirectoryReader().loadData({directoryPath: \"./data2\"})\n",
    "const index2 = await VectorStoreIndex.fromDocuments(documents2)\n",
    "const queryEngine2 = index2.asQueryEngine()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "46868774-6909-4619-9643-b758e37105a2",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "React is a JavaScript library for building user interfaces. It was officially open-sourced in May 2013 at JSConf US, marking a significant turning point in its adoption and growth. React enables the development of native Android, iOS, and Universal Windows Platform applications through React Native. Additionally, React Fiber, a new set of internal algorithms for rendering, was announced by Facebook in April 2017, aiming to improve the performance and efficiency of React's rendering system.\n"
     ]
    }
   ],
   "source": [
    "let response2 = await queryEngine2.query({query:\"What is React?\"})\n",
    "console.log(response2.toString())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "89aed556",
   "metadata": {},
   "source": [
    "### Create a router query engine"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "d04023ce-1551-4618-9dd3-6c157814004f",
   "metadata": {
    "height": 217
   },
   "outputs": [],
   "source": [
    "const queryEngine = await RouterQueryEngine.fromDefaults({\n",
    "  queryEngineTools: [\n",
    "    {\n",
    "      queryEngine: queryEngine1,\n",
    "      description: \"Useful for questions about Dan Abramov\",\n",
    "    },\n",
    "    {\n",
    "      queryEngine: queryEngine2,\n",
    "      description: \"Useful for questions about the React library\",\n",
    "    },\n",
    "  ],\n",
    "});"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "27e763cf",
   "metadata": {},
   "source": [
    "### Ask it a question"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "60917ee5-a48e-4507-a600-84afa5e8f225",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "ename": "Error",
     "evalue": "Got invalid JSON object. Error: SyntaxError: Expected property name or '}' in JSON at position 16 (line 3 column 9). Got JSON string: [\n    {\n        choice: 2,\n        reason: \"Useful for questions about the React library\"\n    }\n]",
     "output_type": "error",
     "traceback": [
      "Stack trace:",
      "Error: Got invalid JSON object. Error: SyntaxError: Expected property name or '}' in JSON at position 16 (line 3 column 9). Got JSON string: [",
      "    {",
      "        choice: 2,",
      "        reason: \"Useful for questions about the React library\"",
      "    }",
      "]",
      "    at SelectionOutputParser.parse (file:///home/jovyan/.cache/deno/npm/registry.npmjs.org/llamaindex/0.1.18/dist/outputParsers/selectors.js:27:23)",
      "    at LLMSingleSelector._select (file:///home/jovyan/.cache/deno/npm/registry.npmjs.org/llamaindex/0.1.18/dist/selectors/llmSelectors.js:105:43)",
      "    at eventLoopTick (ext:core/01_core.js:169:7)",
      "    at async LLMSingleSelector.select (file:///home/jovyan/.cache/deno/npm/registry.npmjs.org/llamaindex/0.1.18/dist/selectors/base.js:23:16)",
      "    at async RouterQueryEngine.queryRoute (file:///home/jovyan/.cache/deno/npm/registry.npmjs.org/llamaindex/0.1.18/dist/engines/query/RouterQueryEngine.js:73:24)",
      "    at async RouterQueryEngine.query (file:///home/jovyan/.cache/deno/npm/registry.npmjs.org/llamaindex/0.1.18/dist/engines/query/RouterQueryEngine.js:64:26)",
      "    at async <anonymous>:1:38"
     ]
    }
   ],
   "source": [
    "let response3 = await queryEngine.query({query: \"What is React?\"})\n",
    "console.log(response3.toString())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "dc142f60-1989-4110-b0c8-1599939f557c",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "In college, Dan Abramov slacked off, skipped lectures, and eventually stopped attending altogether. He also lost motivation for sitting at lectures and doing exercises, leading him to not show up for midterm exams and eventually return his Macbook.\n"
     ]
    }
   ],
   "source": [
    "let response4 = await queryEngine.query({query: \"What did Dan Abramov do in college?\"})\n",
    "console.log(response4.toString())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d4b4002",
   "metadata": {},
   "source": [
    "### Define a simple function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "4a9ebe6c-3d8c-47d0-bdc3-48ccc379f050",
   "metadata": {
    "height": 64
   },
   "outputs": [],
   "source": [
    "function sumNumbers({a,b}) {\n",
    "  return a + b;\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6cebe16",
   "metadata": {},
   "source": [
    "### Explain the function in a JSON structure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "91010aa1-7288-4d9c-a55f-e4538345e414",
   "metadata": {
    "height": 251
   },
   "outputs": [],
   "source": [
    "const sumJSON = {\n",
    "  type: \"object\",\n",
    "  properties: {\n",
    "    a: {\n",
    "      type: \"number\",\n",
    "      description: \"The first number\",\n",
    "    },\n",
    "    b: {\n",
    "      type: \"number\",\n",
    "      description: \"The second number\",\n",
    "    },\n",
    "  },\n",
    "  required: [\"a\", \"b\"],\n",
    "};"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7b88cc1c",
   "metadata": {},
   "source": [
    "### Make the new function into a tool"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "7a1f712a-d24e-4e38-bcf9-57a302f743c1",
   "metadata": {
    "height": 98
   },
   "outputs": [],
   "source": [
    "const sumFunctionTool = new FunctionTool(sumNumbers, {\n",
    "  name: \"sumNumbers\",\n",
    "  description: \"Use this function to sum two numbers\",\n",
    "  parameters: sumJSON,\n",
    "});"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fc946009",
   "metadata": {},
   "source": [
    "### Make the router query engine into a tool"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "abc6c491-43b8-4944-a383-46551a415790",
   "metadata": {
    "height": 132
   },
   "outputs": [],
   "source": [
    "const queryEngineTool = new QueryEngineTool({\n",
    "    queryEngine: queryEngine,\n",
    "    metadata: {\n",
    "        name: \"react_and_dan_abramov_engine\",\n",
    "        description: \"A tool that can answer questions about Dan Abramov and React\",\n",
    "    },\n",
    "});"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf84c156",
   "metadata": {},
   "source": [
    "### Use the two tools"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "0c03fb86-d82e-4ab6-94ef-6b1407a4bb24",
   "metadata": {
    "height": 81
   },
   "outputs": [],
   "source": [
    "const agent = new OpenAIAgent({\n",
    "    tools: [queryEngineTool, sumFunctionTool],\n",
    "    verbose: true\n",
    "})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f1221ee1",
   "metadata": {},
   "source": [
    "### Let's ask it a question!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "70a427c6-3e0e-4216-8824-f229147e4209",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Added user message to memory!: What is React? Use a tool.\n",
      "=== Calling Function ===\n",
      "Calling function: react_and_dan_abramov_engine with args: {\n",
      "  \"query\": \"What is React?\"\n",
      "}\n",
      "Got output React is a JavaScript library for building user interfaces. It was officially open-sourced in May 2013 at JSConf US, marking a significant turning point in its adoption and growth. React enables the development of native Android, iOS, and Universal Windows Platform applications through React Native. Additionally, React Fiber, a new set of internal algorithms for rendering, was announced by Facebook in April 2017, aiming to improve performance and feature development of the React library. The syntax for programming with React remains the same, with only the execution method changing.\n",
      "==========================\n",
      "React is a JavaScript library for building user interfaces. It was officially open-sourced in May 2013 and has since gained significant popularity. React allows developers to create reusable UI components and efficiently update and render them when the underlying data changes. It uses a virtual DOM (Document Object Model) to efficiently update only the necessary parts of the UI, resulting in faster and more efficient rendering. React is often used in combination with other libraries or frameworks, such as Redux, to manage application state.\n"
     ]
    }
   ],
   "source": [
    "let response5 = await agent.chat({message:\"What is React? Use a tool.\"})\n",
    "console.log(response5.toString())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "c3460d6a-fbc0-4d93-8a79-529756fec52d",
   "metadata": {
    "height": 47
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Added user message to memory!: What is 501 + 5?\n",
      "The sum of 501 and 5 is 506.\n"
     ]
    }
   ],
   "source": [
    "let response6 = await agent.chat({message:\"What is 501 + 5?\"})\n",
    "console.log(response6.toString())"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Deno",
   "language": "typescript",
   "name": "deno"
  },
  "language_info": {
   "file_extension": ".ts",
   "mimetype": "text/x.typescript",
   "name": "typescript",
   "nb_converter": "script",
   "pygments_lexer": "typescript",
   "version": "5.3.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}