{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# šŸš€ Day 1: Guided Lab - Your First LLM Interactions with Gemini\n", "\n", "Welcome to your first hands-on session with Large Language Models!\n", "\n", "## Learning Objectives\n", "By the end of this lab, you will be able to:\n", "- āœ… Set up and use the Gemini API in Google Colab\n", "- āœ… Make API calls and understand the response structure\n", "- āœ… Experiment with parameters like temperature and max tokens\n", "- āœ… Build multi-turn conversations\n", "- āœ… Create a simple business application\n", "\n", "## Time: 90 minutes\n", "\n", "---\n", "\n", "## šŸŽÆ Ground Rules for Working with LLMs\n", "\n", "Keep these principles in mind throughout the course:\n", "\n", "1. **If unsure, say so** - LLMs should acknowledge uncertainty, not make things up\n", "2. **Output format matters** - Always specify the format you need (especially JSON)\n", "3. **Don't invent facts** - LLMs should only use information provided in the input\n", "4. **Verify important outputs** - Always spot-check critical results\n", "\n", "---" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Part 1: Setup (15 minutes)\n", "\n", "### 1.1 Install the Gemini SDK\n", "\n", "Run the cell below to install the Google GenAI package.\n", "\n", "> āš ļø You may see a dependency warning about `google-auth` - this is safe to ignore." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!pip -q install -U google-genai" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 1.2 Load Your API Key\n", "\n", "Make sure you've added your `GEMINI_API_KEY` to Colab Secrets (šŸ”‘ icon in the left sidebar).\n", "\n", "The secret name must be exactly: `GEMINI_API_KEY`" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import os\n", "import json\n", "import time\n", "from datetime import datetime, timezone\n", "from google import genai\n", "from google.genai import types\n", "\n", "# --- Load API key ---\n", "# Option A (recommended): Colab Secrets\n", "try:\n", " from google.colab import userdata\n", " os.environ[\"GEMINI_API_KEY\"] = userdata.get(\"GEMINI_API_KEY\")\n", "except Exception:\n", " pass\n", "\n", "# Option B (fallback): hidden paste\n", "if not os.environ.get(\"GEMINI_API_KEY\"):\n", " import getpass\n", " os.environ[\"GEMINI_API_KEY\"] = getpass.getpass(\"Paste your GEMINI_API_KEY (input hidden): \")\n", "\n", "# Verify\n", "if os.environ.get(\"GEMINI_API_KEY\"):\n", " print(\"āœ… API key loaded successfully!\")\n", "else:\n", " print(\"āŒ API key not found. Check your Colab Secrets.\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 1.3 Initialize Client and Helper Functions\n", "\n", "We'll set up:\n", "1. The Gemini client\n", "2. A **prompt log** to track all our API calls (useful for learning!)\n", "3. Helper functions we'll use throughout the course" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Initialize the client\n", "client = genai.Client()\n", "\n", "# Model configuration\n", "MODEL_ID = \"gemini-2.0-flash-lite\" # Fast and free-tier friendly\n", "\n", "# --- Prompt Logging ---\n", "# This helps you track what you tried and what worked\n", "PROMPT_LOG = []\n", "\n", "def _now():\n", " \"\"\"Get current timestamp.\"\"\"\n", " return datetime.now(timezone.utc).isoformat().replace('+00:00', 'Z')\n", "\n", "def generate(prompt, temperature=0.7, max_tokens=500, log=True):\n", " \"\"\"\n", " Generate text using Gemini with customizable parameters.\n", " Automatically logs all calls for later review.\n", "\n", " Args:\n", " prompt (str): The input prompt\n", " temperature (float): Creativity level (0.0 to 2.0)\n", " max_tokens (int): Maximum output length\n", " log (bool): Whether to log this call\n", "\n", " Returns:\n", " str: The generated text\n", " \"\"\"\n", " t0 = time.time()\n", "\n", " response = client.models.generate_content(\n", " model=MODEL_ID,\n", " contents=prompt,\n", " config=types.GenerateContentConfig(\n", " temperature=temperature,\n", " max_output_tokens=max_tokens\n", " )\n", " )\n", "\n", " text = response.text or \"\"\n", " latency = time.time() - t0\n", "\n", " # Log the call\n", " if log:\n", " PROMPT_LOG.append({\n", " \"ts\": _now(),\n", " \"model\": MODEL_ID,\n", " \"temperature\": temperature,\n", " \"max_tokens\": max_tokens,\n", " \"prompt\": prompt,\n", " \"output\": text,\n", " \"latency_s\": round(latency, 3)\n", " })\n", "\n", " return text\n", "\n", "def try_parse_json(text):\n", " \"\"\"\n", " Attempt to parse JSON from LLM output.\n", " Returns (success: bool, parsed_data: dict/list or None)\n", " \"\"\"\n", " # Try direct parse\n", " try:\n", " return True, json.loads(text)\n", " except json.JSONDecodeError:\n", " pass\n", "\n", " # Try to extract JSON from markdown code blocks\n", " import re\n", " json_match = re.search(r'```(?:json)?\\s*([\\s\\S]*?)```', text)\n", " if json_match:\n", " try:\n", " return True, json.loads(json_match.group(1))\n", " except json.JSONDecodeError:\n", " pass\n", "\n", " # Try to find JSON object/array in text\n", " json_match = re.search(r'[\\[\\{][\\s\\S]*[\\]\\}]', text)\n", " if json_match:\n", " try:\n", " return True, json.loads(json_match.group())\n", " except json.JSONDecodeError:\n", " pass\n", "\n", " return False, None\n", "\n", "def show_log(n=3):\n", " \"\"\"Display the last n entries from the prompt log.\"\"\"\n", " for row in PROMPT_LOG[-n:]:\n", " print(\"=\" * 70)\n", " print(f\"Time: {row['ts']} | Temp: {row['temperature']} | Latency: {row['latency_s']}s\")\n", " print(\"--- PROMPT (first 500 chars) ---\")\n", " print(row[\"prompt\"][:500])\n", " print(\"--- OUTPUT (first 500 chars) ---\")\n", " print(row[\"output\"][:500])\n", "\n", "print(f\"āœ… Client initialized! Using model: {MODEL_ID}\")\n", "print(f\"šŸ“ Prompt logging enabled. Use show_log() to review your calls.\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 1.4 Test Your Connection\n", "\n", "Let's make sure everything works with a simple test." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Simple test call\n", "response = generate(\n", " \"Say 'Hello, I am ready to learn!' in a friendly way.\",\n", " temperature=0.7\n", ")\n", "\n", "print(response)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "šŸŽ‰ **If you see a friendly greeting above, you're all set!**\n", "\n", "---" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Part 2: Basic Text Generation (15 minutes)\n", "\n", "### 2.1 Your First Business Prompt\n", "\n", "Let's generate a product description - a common business use case." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Simple product description\n", "prompt = \"Write a product description for wireless noise-canceling headphones in 2-3 sentences.\"\n", "\n", "response = generate(prompt)\n", "\n", "print(\"Generated Description:\")\n", "print(response)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 2.2 Run It Multiple Times\n", "\n", "Run the cell below multiple times. Notice how the output varies each time!\n", "\n", "This is because LLMs are **probabilistic** - they don't always give the same answer." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Run this cell 3-4 times and observe the different outputs\n", "response = generate(\n", " \"Suggest a creative name for a new coffee shop.\",\n", " temperature=0.9 # Higher temperature = more variety\n", ")\n", "\n", "print(response)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 2.3 Understanding the Response Object\n", "\n", "Let's look at more details from the API response, including token usage." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Make a call and examine the full response\n", "full_response = client.models.generate_content(\n", " model=MODEL_ID,\n", " contents=\"What are three benefits of cloud computing for small businesses?\"\n", ")\n", "\n", "# The main text output\n", "print(\"=== Response Text ===\")\n", "print(full_response.text)\n", "\n", "# Usage statistics (tokens used)\n", "print(\"\\n=== Usage Statistics ===\")\n", "print(f\"Input tokens: {full_response.usage_metadata.prompt_token_count}\")\n", "print(f\"Output tokens: {full_response.usage_metadata.candidates_token_count}\")\n", "print(f\"Total tokens: {full_response.usage_metadata.total_token_count}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "šŸ’” **Why tokens matter:**\n", "- API pricing is based on tokens\n", "- There are limits on input and output tokens\n", "- 1 token ā‰ˆ 4 characters or Ā¾ of a word in English\n", "\n", "---" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Part 3: Exploring Parameters (15 minutes)\n", "\n", "### 3.1 Temperature: Controlling Creativity\n", "\n", "**Temperature** controls how \"creative\" or \"random\" the output is:\n", "- `temperature=0.0` ā†’ Deterministic, consistent outputs\n", "- `temperature=1.0` ā†’ More creative, varied outputs\n", "- `temperature=2.0` ā†’ Very random (often too chaotic)\n", "\n", "Let's see this in action!" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "prompt = \"Suggest a name for a new smartphone app that helps people track their fitness goals.\"\n", "\n", "print(\"=\" * 50)\n", "print(\"TEMPERATURE = 0.0 (Deterministic)\")\n", "print(\"Running 3 times - outputs should be very similar:\")\n", "print(\"=\" * 50)\n", "\n", "for i in range(3):\n", " response = generate(prompt, temperature=0.0)\n", " print(f\"Run {i+1}: {response.strip()[:300]}\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(\"=\" * 50)\n", "print(\"TEMPERATURE = 1.0 (Creative)\")\n", "print(\"Running 3 times - outputs should be different:\")\n", "print(\"=\" * 50)\n", "\n", "for i in range(3):\n", " response = generate(prompt, temperature=1.0)\n", " print(f\"Run {i+1}: {response.strip()[:300]}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### šŸ’­ Discussion Question\n", "\n", "**When would you use low temperature vs high temperature in a business context?**\n", "\n", "- Low temperature (0.0-0.3): _______________________\n", "- High temperature (0.7-1.0): _______________________" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 3.2 Max Output Tokens: Controlling Length\n", "\n", "You can limit how long the response should be." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "prompt = \"Explain the concept of machine learning to a business executive.\"\n", "\n", "# Short response (max 50 tokens)\n", "print(\"=== SHORT (max 50 tokens) ===\")\n", "response_short = generate(prompt, max_tokens=50)\n", "print(response_short)\n", "\n", "# Check actual length from log\n", "print(f\"\\n(Check the log for actual token count)\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Longer response (max 200 tokens)\n", "print(\"=== LONGER (max 200 tokens) ===\")\n", "response_long = generate(prompt, max_tokens=200)\n", "print(response_long)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 3.3 Review Your Prompt Log\n", "\n", "Let's see what we've tried so far!" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# See how many calls we've made\n", "print(f\"Total API calls so far: {len(PROMPT_LOG)}\")\n", "\n", "# Show the last 3 calls\n", "show_log(3)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "---\n", "\n", "## Part 4: Multi-turn Conversations (15 minutes)\n", "\n", "LLMs can maintain context across multiple exchanges, just like a real conversation.\n", "\n", "### 4.1 Starting a Chat Session" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Create a chat session\n", "chat = client.chats.create(model=MODEL_ID)\n", "\n", "# First message\n", "response1 = chat.send_message(\"Hi! I'm looking for a laptop for video editing. What should I consider?\")\n", "print(\"User: Hi! I'm looking for a laptop for video editing. What should I consider?\")\n", "print(f\"\\nAssistant: {response1.text}\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Follow-up message - the model remembers the context!\n", "response2 = chat.send_message(\"My budget is around $1500. Any specific recommendations?\")\n", "print(\"User: My budget is around $1500. Any specific recommendations?\")\n", "print(f\"\\nAssistant: {response2.text}\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Another follow-up\n", "response3 = chat.send_message(\"What about battery life? I travel a lot.\")\n", "print(\"User: What about battery life? I travel a lot.\")\n", "print(f\"\\nAssistant: {response3.text}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 4.2 View Chat History\n", "\n", "You can access the full conversation history." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# View the conversation history\n", "print(\"=== Full Chat History ===\")\n", "for i, message in enumerate(chat._curated_history):\n", " role = \"šŸ‘¤ User\" if message.role == \"user\" else \"šŸ¤– Assistant\"\n", " text = message.parts[0].text if message.parts else \"\"\n", " preview = text[:150] + \"...\" if len(text) > 150 else text\n", " print(f\"\\n{role}: {preview}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### šŸ’­ Discussion Question\n", "\n", "**How might multi-turn conversations be useful in a business application?**\n", "\n", "Think of 2-3 examples: _______________________\n", "\n", "---" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Part 5: Business Application - Email Response Generator (15 minutes)\n", "\n", "Let's build something practical: a customer service email response generator.\n", "\n", "### 5.1 Basic Email Response" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def generate_email_response(customer_email, company_name=\"TechCorp\", tone=\"professional\"):\n", " \"\"\"\n", " Generate a customer service response to an email.\n", "\n", " Args:\n", " customer_email (str): The customer's email content\n", " company_name (str): Your company name\n", " tone (str): Response tone (professional, friendly, formal)\n", "\n", " Returns:\n", " str: Generated email response\n", " \"\"\"\n", " prompt = f\"\"\"You are a customer service representative for {company_name}.\n", "Write a {tone} response to this customer email.\n", "\n", "Customer Email:\n", "---\n", "{customer_email}\n", "---\n", "\n", "Requirements:\n", "- Acknowledge their concern empathetically\n", "- Provide a clear solution or next steps\n", "- Keep the response under 150 words\n", "- End with a positive note\n", "- Do not invent facts not present in the email\n", "\n", "Response:\"\"\"\n", "\n", " return generate(prompt, temperature=0.3) # Low temperature for consistency\n", "\n", "\n", "# Test with a sample email\n", "sample_email = \"\"\"\n", "Hi,\n", "\n", "I ordered a laptop last week (Order #12345) and it arrived today with a cracked screen.\n", "This is really frustrating as I needed it for an important presentation tomorrow.\n", "\n", "What can you do about this?\n", "\n", "Thanks,\n", "John\n", "\"\"\"\n", "\n", "response = generate_email_response(sample_email)\n", "print(\"Generated Response:\")\n", "print(response)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 5.2 Try Different Tones" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Same email, different tones\n", "print(\"=== PROFESSIONAL TONE ===\")\n", "print(generate_email_response(sample_email, tone=\"professional\"))\n", "\n", "print(\"\\n\" + \"=\"*50 + \"\\n\")\n", "\n", "print(\"=== FRIENDLY TONE ===\")\n", "print(generate_email_response(sample_email, tone=\"friendly and warm\"))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 5.3 Meeting Summary Generator\n", "\n", "Another common business use case: summarizing meeting notes." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def summarize_meeting(transcript):\n", " \"\"\"\n", " Summarize a meeting transcript into key points and action items.\n", " \"\"\"\n", " prompt = f\"\"\"Summarize the following meeting transcript.\n", "\n", "Provide:\n", "1. A 2-3 sentence summary\n", "2. Key decisions made (bullet points)\n", "3. Action items with owners (if mentioned)\n", "\n", "Rules:\n", "- Only include information from the transcript\n", "- If owner/deadline not mentioned, write \"Not specified\"\n", "- Be concise\n", "\n", "Meeting Transcript:\n", "---\n", "{transcript}\n", "---\n", "\n", "Summary:\"\"\"\n", "\n", " return generate(prompt, temperature=0.2)\n", "\n", "\n", "# Sample meeting transcript\n", "meeting_transcript = \"\"\"\n", "Sarah: Okay everyone, let's discuss the Q2 marketing budget.\n", "\n", "Mike: I think we should increase social media spending by 20%. Our Instagram campaigns\n", "have been performing really well.\n", "\n", "Sarah: Good point. What about the trade show in March?\n", "\n", "Lisa: The trade show costs about $15,000. I think it's worth it for the leads we get.\n", "Last year we got 50 qualified leads from it.\n", "\n", "Mike: Agreed. Let's keep the trade show budget.\n", "\n", "Sarah: Alright, so we'll increase social media by 20% and maintain trade show spending.\n", "Mike, can you prepare a detailed social media plan by Friday?\n", "\n", "Mike: Sure, I'll have it ready.\n", "\n", "Sarah: Lisa, please confirm our trade show registration.\n", "\n", "Lisa: Will do, I'll handle it this week.\n", "\n", "Sarah: Great, let's reconvene next Tuesday to finalize everything.\n", "\"\"\"\n", "\n", "summary = summarize_meeting(meeting_transcript)\n", "print(\"Meeting Summary:\")\n", "print(summary)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 5.4 Testing JSON Output (Preview of Day 2)\n", "\n", "Sometimes you need structured data, not just text. Let's try getting JSON output." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Request JSON output\n", "json_prompt = \"\"\"Extract action items from this meeting note. Return JSON ONLY.\n", "\n", "Meeting note: \"Mike will prepare social media plan by Friday. Lisa will confirm trade show registration this week.\"\n", "\n", "Schema:\n", "[\n", " {\"owner\": \"name\", \"task\": \"description\", \"deadline\": \"when\"}\n", "]\n", "\n", "JSON:\"\"\"\n", "\n", "response = generate(json_prompt, temperature=0.0)\n", "print(\"Raw output:\")\n", "print(response)\n", "\n", "# Try to parse it\n", "success, parsed = try_parse_json(response)\n", "print(f\"\\nValid JSON? {success}\")\n", "if success:\n", " print(\"Parsed data:\")\n", " for item in parsed:\n", " print(f\" - {item}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "---\n", "\n", "## Part 6: Wrap-up (15 minutes)\n", "\n", "### 6.1 Export Your Prompt Log\n", "\n", "Save a record of everything you tried today!" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "\n", "# Convert log to DataFrame\n", "df = pd.DataFrame(PROMPT_LOG)\n", "\n", "# Save to CSV\n", "df.to_csv(\"day1_guided_lab_prompt_log.csv\", index=False)\n", "print(f\"āœ… Saved {len(df)} prompts to: day1_guided_lab_prompt_log.csv\")\n", "\n", "# Show summary\n", "print(f\"\\nšŸ“Š Session Summary:\")\n", "print(f\" Total API calls: {len(df)}\")\n", "print(f\" Average latency: {df['latency_s'].mean():.2f}s\")\n", "print(f\" Temperature range: {df['temperature'].min()} - {df['temperature'].max()}\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### 6.2 Key Takeaways\n", "\n", "**What we learned today:**\n", "\n", "1. **API Basics**: You can interact with LLMs through simple API calls\n", "2. **Parameters Matter**: Temperature and max tokens control output behavior\n", "3. **Conversations**: LLMs can maintain context across multiple turns\n", "4. **Business Value**: Even simple prompts can automate useful tasks\n", "5. **Logging Helps**: Tracking your prompts helps you learn what works\n", "\n", "### 6.3 Reflection Questions\n", "\n", "Before moving to the independent lab, discuss with a partner:\n", "\n", "1. What surprised you most about working with LLMs?\n", "2. What limitations did you notice?\n", "3. What business process in your experience could benefit from this technology?\n", "\n", "---\n", "\n", "## šŸŽÆ Ready for the Independent Lab!\n", "\n", "In the next session, you'll work independently (or with a partner) to:\n", "- Build a content generation suite\n", "- Create text transformation tools\n", "- Develop an interactive Q&A system\n", "\n", "**Save this notebook** - you'll reuse the helper functions!" ] } ], "metadata": { "colab": { "provenance": [], "toc_visible": true }, "kernelspec": { "display_name": "Python 3", "name": "python3" }, "language_info": { "name": "python" } }, "nbformat": 4, "nbformat_minor": 0 }