Model HQ Documentation

The Challenge

It's a sweltering summer in the NYC area — home base for Model HQ — and like many, we recently ran into an unexpected crisis: our AC unit gave out.

No big deal, right? Just call in a tech for a quick fix.

But here's the twist: ❌ the unit wasn't repairable. It was running on R22, a refrigerant banned in the U.S. for years due to its environmental impact.

We were stunned — we'd been faithfully maintaining this system with regular servicing. Why didn't anyone tell us sooner?

That's when it clicked. This isn't just our problem — it's a nationwide issue affecting millions of legacy AC units. So, we built a fix of what we would do if we had an AC company. 🔧

And it's not just for HVAC…

This "photo-to-email" workflow can be easily adapted across many industries and use cases:

Manufacturing: Serial number tracking and part recall notifications

Auto service: VIN plate photos triggering service reminders

Medical devices: Compliance alerts based on equipment ID plates

Appliances & home systems: Warranty, upgrade, or service eligibility detection

One image. One workflow. Endless automation.

Snap. Analyze. Email.

With Model HQ, we built an automated agent workflow that takes:

A single photo of an AC unit's side panel (a.k.a. the data plate)

Extracts the model number, and cross checks this against the serial, refrigerant type, year, and manufacturer (this part is dummy data – insert your own relevant data here for other uses)

Cross-checks technical documents like the R22 phaseout guide

Identifies at-risk units

And then crafts and sends a personalized customer service + marketing email

—all in seconds.

Why It Matters

This workflow doesn't just automate technician follow-ups. It creates instant, data-driven customer engagement — and a seamless path to value-added services.

📣 Free marketing ideas, anyone? Because this one just turned field service into a sales engine.

Step-by-Step Process

Download the Image and Files you will Need to Run this Demo

Download a clear photo of the AC unit sticker showing the model number (save the photo here to follow along with this tutorial)

Download the AC Manufacturer, Serial Number, Year of Manufacture and Freon type information .CSV called AC Unit Data you will need to create a look-up database (need to provide access to file)

Download the R22 Technical Document the agent will reference for information.

Find all the required images and documents here.

Inputting the Correct Gmail Credentials

Make sure that you have filled in the Sender Email Address and Sender Email App Access Password (this is different than your normal password) in Model HQ in:

Configs ⚙️ in the upper right hand corner then Credentials then Sender Email Address (GMAIL) and Sender Email Password (App Access)

If you don't have an App Access Password, you must enable 2-Factor Authentication for your Gmail account, and click on "Create and manage your app passwords" under this website:

https://support.google.com/mail/answer/185833?hl=en

Building the Agent, Starting with Specifying the Right File Input

Start Building a New Agent by going to Agent > Build New > Start Building > then name this new Agent AC Marketing

As with all Agent Process, start by specifying the File Inputs from the User. Here, we will ask the AC Tech to input a photo of the Data Panel (the photo we downloaded above), and provide the name and email address of the customer (the AC tech will manually enter this info for this demo).

1Go to '+ Inputs'

2Select for Input Type, Image File Input from the drop down menu, then add a Description that will help the user understand what the input should be "Insert photo of the AC Side Panel" then select ' >'

3Next select '+' to add another Input type – 'Text Input – standard 'chat-style' text box' and add a Description "Input the Name and Email Address of customer"

4Once you see both Input Types with blue check marks next to them, click Save and Exit

Using the Vision Agent to auto-extract the model number from the panel image

Row 1:

From the Service Menu, select Vision to processes the image.

Enter in Instruction: What is the model number?

Choose for Context: User Image

Vision agent extracts the text and pulls the model number from the sticker image.

Using an extract Agent to extract a clean model number

You may find that the Vision Model can be extra-chatty and volunteers too much information. This is where an Extract Agent comes in handy to read through the chatty text the Vision Model produces to get exactly the information we need to be able to automatically query our dataset using our Text to SQL agent.

Row 2:

From the Service Menu, Select Extract to get a clean model number without any surrounding text

Enter in Instruction: the model number

Choose for Context: Vision_1 (We are asking the Extract agent to pull the information gathered from the Vision agent in Step 1).

Extract agent pulls only the information needed to perform the SQL query

Adding a Built-In Database to the Agent Workflow

This is where we build a table of the .CSV file for the query. Not only that, we will have this .CSV file BUILT-IN to the process as a permanent source so that the user does not have to input this source each time this process is run.

To enable this built-in data-source feature, click on '+Files' at the bottom of the screen.

Select 'table' as the file type and find the .CSV file you downloaded at Step 1 of this tutorial in the Choose file bar.

Once the file is selected, click '+'

We will also use the R-22 Technical Information file as a permanent data source, so we will add it in this step.

In File Type, select 'document' (it is usually pre-selected as the first option)

Then find the R-22 Technical document file you downloaded at Step 1 of this tutorial in the Choose File bar.

Once the file is selected, click '+'

Note: Once all relevant content files are selected (in our case, the .CSV file with Manufacturer Name and the Document file with the R-22 Technical information), Click 'Save + Exit'. It is important to confirm your selection by clicking 'Save + Exit' at this step. Simply pressing the Home button may cause you to lose the files you just uploaded.

Building a Table of AC Unit Data

Now that we have specified the built-in databases for this workflow, let's start building our database table.

Row 3:

From the Services Menu, select 'Build_Table' to automatically build a queryable database table from the .CSV

Enter in Instruction: ac_data

Choose for Context: Updated_AC_Unit_Data

This builds the table that we can now start to query in natural language

Query AC Unit Database

Row 4:

From the Services Menu, select 'Query_custom_table'

Enter in Instruction (must enter exactly): What is the serial number, manufacturer, year of manufacture for {{the_model_number}}

Choose for Context: ac_data

This introduces the concept of a using a variable in our query. Using the model number we gathered in Row 2 from the Extract agent, this allows us to query in natural language information about an earlier search.

Row 5:

From the Services Menu, select 'Query_custom_table'

Enter in Instruction (must enter exactly): What is freon type for {{the_model_number}}

Choose for Context: ac_data

This continues the concept of a using a variable in our query. Using the model number we gathered in Row 2 from the Extract agent, this allows us to query in natural language information about an earlier search.

By performing these actions, the model number is used to query the AC unit table to find:

Serial number
Manufacturer
Year of manufacture

A second query checks:

Freon type (e.g., R22)

Contextual RAG Lookup

To find information on R22, we will next ask the agent parses the document: R22 Technical Documentation.pdf'

Row 6:

From the Services Menu, select 'Parse_document'

Enter in Instruction (must enter exactly): R22 technical doc

Choose for Context: R22_Technical Document

This step parses the document that is in a PDF format, and normalizes the text and prepares the text with auto-chunking and indexing for semantic (natural language) queries

Row 7:

Next we can start to ask natural language questions

From the Services Menu, select 'Rag_answer'

Enter in Instruction (must enter exactly): What is R22 and is it still being used?

Choose for Context: R22_Technical Document

This step is now using the document to pull information directly from the source to answer questions in natural language RAG model answers: "What is R22 and is it still being used?"

Match the Customer Info

From the field tech's input (e.g., "Today we visited customer Jane Smith...and email address"): The agent extracts customer name and email address

Row 8:

From the Services Menu, select 'Extract'

Enter in Instruction (must enter exactly): customer name

Choose for Context: Main Input

Row 9:

From the Services Menu, select 'Extract Tiny'

Enter in Instruction (must enter exactly): email address

Choose for Context: Main Input

These steps will extract the customer's name and email address – the key pieces of information we need to send an email.

Chat Agent Drafts the Email

Using a chat model, the system generates a customer-ready email:

Personalized with:

Customer name
Model + Serial Number
Year + Manufacturer
Notes if R22 is used
Suggests sales outreach if replacement is recommended

Row 10:

From the Services Menu, select 'Chat'

Enter in Instruction (for your first use, we highly recommend you copy and paste the instruction. Afterwards, feel free to experiment but we have found this to work well – it is key to use the variables correctly):

Choose for Context: Vision 1

Email Draft Instruction

You are a customer support agent named Amanda who works for a heating and cooling company named Model AC. Write a nice email to a customer named {{customer_name}} about our service delivery that took place today with our technician David. Using the information provided here about {{the_model_number}}, the {{query_custom_table_4}}, the {{query_custom_table_5}} and {{rag_answer_7}}, reference the exact unit, including serial number, model, manufacture and year of manufacture of the AC unit which is information that is found as follows: {{query_custom_table_4}}. If the unit uses R22, mention to the customer that we highly recommend replacing the unit and that the Sales Rep will contact them with a price quote on a great new system. Do not use brackets or have blanks in the email. If you can't fill in the information, don't discuss it.

This creates a nice email from the chatbot – however, these chatbots come with intros about the steps they took to create the email, so the next step is also necessary.

Row 11:

From the Services Menu, select 'Chat'

Enter in Instruction (copy and paste the first time then experiment after):

Choose for Context: Chat_11

Email Formatting Instruction

You are an email formatting expert. Review {{Chat_10}} and format the email nicely. Do not included any responses or any other text in this email other than the context. Do not add "here is" or any other responses whatsoever. This will cause the outcome to fail.

A second chat step formats the email:

Clean paragraphs
No headers or extra labels
Pure copy-paste ready

Send the Email

Model HQ auto-sends the formatted email to the retrieved email address using the send_email node.

Row 12:

From the Services Menu, select 'Send_email'

Enter in Instruction (must enter exactly): {{email_address}}

Choose for Context: Chat_11

Now that you entered all 12 rows, select '>'. This is the screenshot of what you should see:

Running the Workflow

You did it~~~ Now let's test by selecting 'Run' then 'Action' >

Main Input:

For the Text input in Main Input, enter your name and email address so you can test: [Your Name] [email]

For the Upload Image File – AC Marketing, choose the image file of the AC panel you downloaded in Step 1 and select >.

You will see the following results

Recap Summary: What This Agent Does

Task	Output
🖼️ Image Input	Extracted AC model number
📊 Table Query	Unit specs + refrigerant type
📄 Document RAG	Info on R22
🙍‍♀️ Customer Lookup	Name + email address
📨 Email Generation	Personalized & well-formatted
🚀 Final Step	Sends the email automatically

🖼️ Image Input

Extracted AC model number

📊 Table Query

Unit specs + refrigerant type

📄 Document RAG

Info on R22

🙍‍♀️ Customer Lookup

Name + email address

📨 Email Generation

Personalized & well-formatted

🚀 Final Step

Sends the email automatically

Need Help?

If you encounter any issues while setting up this workflow, feel free to contact our support team at support@aibloks.com

From Data Plate to Inbox: Turn a Single Photo into a Powerful Customer Email with Model HQ