Draft AI-powered Email Replies Using Hugging Face

Get Your Digital Key (API Token)

First, let's get your unique digital key, known as an API token. Think of an API (Application Programming Interface) as a waiter in a restaurant. You give your order (request) to the waiter, and they take it to the kitchen (the AI model). Your API token is like your VIP pass that tells the waiter you're allowed to place an order.

1. Log in to your Hugging Face account.
2. Click on your profile picture in the top right corner.
3. From the dropdown menu, select "Settings".
4. On the left-hand side, click "Access Tokens".
5. Click the button that says "New token".
6. Give your token a descriptive name (e.g., "Email-Reply-Bot").
7. Set the Role to "read" for now (you're only "reading" the AI's output).
8. Click "Generate a token".
9. Copy this token immediately! Treat it like a password. You won't see it again once you leave this page. Save it somewhere safe, like a temporary note, as you'll need it soon.

Find an AI "Chef" (Model) for Your Email

Now you need to choose an AI model, which is like selecting a specific chef for your meal. Each model is trained for different tasks. For drafting emails, you'll want a "text generation" model.

1. Go to the Hugging Face Hub.
2. In the search bar, type text-generation.
3. Look for a model that seems suitable and generally popular. For a beginner, gpt2 is a good choice as it's common and effective for general text. Other options might be distilgpt2 (a smaller, faster version) or microsoft/DialoGPT-medium.
4. Once you click on a model (e.g., gpt2), you'll see its Model ID at the top (e.g., gpt2). Make a note of this.

Prepare Your "Order" (API Request)

Now you'll craft the specific instruction you want to send to the AI. This is called a prompt. You'll combine your API token, the model ID, and your prompt into a command. We'll use a tool called curl (pronounced "curl"), which is a common way to send requests from your computer's terminal.

Open your computer's terminal (on Windows, search for "Command Prompt" or "PowerShell"; on Mac/Linux, search for "Terminal").

Here’s the basic structure for your curl command. Remember to replace the placeholders:

curl -X POST \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "inputs": "Draft a polite reply to a customer email saying we will look into their issue.", "parameters": { "max_new_tokens": 50, "return_full_text": false } }' \
  https://api-inference.huggingface.co/models/YOUR_MODEL_ID

Let's break down the important parts:

• YOUR_API_TOKEN: Replace this with the token you copied in Step 1.
• YOUR_MODEL_ID: Replace this with the Model ID you chose in Step 2 (e.g., gpt2).
• "inputs": This is your prompt – the actual text you're asking the AI to respond to.
• "parameters": These are like special instructions for the AI.
  • "max_new_tokens": 50: This tells the AI to generate a reply of approximately 50 new pieces of text (tokens). A token is usually about four characters long. You can adjust this number.
  • "return_full_text": false: This ensures the AI only returns the new text it generated, not your original prompt as well.

curl -X POST \
  -H "Authorization: Bearer hf_YOURTOKEN123" \
  -H "Content-Type: application/json" \
  -d '{ "inputs": "The customer is unhappy with the service and wants a refund. Draft a polite and empathetic reply.", "parameters": { "max_new_tokens": 80, "return_full_text": false } }' \
  https://api-inference.huggingface.co/models/gpt2

Send Your Order to the AI "Kitchen"

Paste the entire curl command (with your actual token and model ID) into your terminal and press Enter.

The AI will process your request (this is called inference). It might take a few seconds, especially the first time you use a specific model, as Hugging Face "loads" the model for you.

Read the AI's "Response" (Draft Email)

After a short wait, you'll see some text appear in your terminal. This will be in a format called JSON (JavaScript Object Notation), which is a common, organised way for programs to send data.

It will look something like this (though the exact text will vary):

[
  {
    "generated_text": "Dear [Customer Name],\n\nThank you for reaching out to us. We understand your frustration regarding the service you received and sincerely apologise for any inconvenience this may have caused. We are committed to ensuring our customers are happy, and we take your feedback seriously. We will investigate this matter thoroughly and get back to you as soon as possible with a resolution.\n\nSincerely,\n[Your Name]"
  }
]

Look for the text inside "generated_text". This is your AI-drafted email reply! You can now copy and paste this text, then customise it with names, specific details, and your brand's voice.