Mi Integration

Accelerating auto part purchase orders with Agentic AI

The context

Mi Integration is a Canadian-founded company specializing in thermoplastic injection molding and engineering for the automotive industry. From seals and lighting elements to HVAC parts and sleek interior trim, they design, engineer, and manufacture complex plastic components that meet the high standards of modern vehicle production.

Their operations in Canada and Mexico support some of the biggest names in the sector, so efficiency is critical. Their supply chain team processes purchase orders that come in as PDF files, each packed with a high number of individual line items and essential details that flow into departments like planning, finance, and order management. But until recently, that process started in a centralized inbox and relied entirely on manual entry.

Mi Integration came to us seeking a faster and more accurate purchase order workflow that was less dependent on human effort. Recognizing the project as a perfect use case for automation to shine, here’s how we used agentic AI to help Mi Integration reclaim time, reduce errors, and let their supply chain team focus on more strategic work (instead of repetitive data wrangling).

Project details

Industry

Manufacturing

Technologies

Gemini Pro

Azure

AWS

GitHub

Outlook

OneDrive

Services

The challenge

Our central challenge was to design a solution flexible enough to handle the messy realities of day-to-day operations, like purchase orders arriving as multi-page PDF attachments filled with part numbers, quantities, pricing, delivery terms, and production timelines. Previously, a supply chain manager had to read through each one manually and transfer the details into a CSV file used by multiple departments. It was time-consuming, repetitive, and left too much room for human error.

An early technical consideration for the project was file format variability. While most purchase orders came in as PDFs, some were sent as CSVs. We had to decide whether to build a system that handled every format or streamline the process around a single one. Prioritizing reliability, we chose PDFs and added upstream conversion steps where needed. For example, any CSV file that came into the inbox would be automatically converted into a PDF.

When it came time to select our model, processing speed was a key consideration. Large language models with strong reasoning capabilities are ideal for interpreting complex documents like purchase orders, though that level of accuracy can slow performance. During peak times, especially early mornings when the supply chain inbox fills up, the system risks becoming overloaded. We had to strike a balance. Rather than switch to a faster but less precise model, we opted to keep the high reasoning accuracy and shape the solution around it. That meant designing the workflow to pace requests, manage volume intelligently, and recover smoothly under pressure.

Engineer working in an automative factory.

The task at hand

This was a short, focused build with no need for a full Discovery Phase. Mi Integration came in clear on what they wanted, so we got to work right away. From the start, they emphasized scalability and ownership. They wanted a solution their team could maintain and grow. It had to be transparent and easy to build on, so getting the architecture right was essential.

We collaborated closely with both their supply chain manager and internal IT lead. The supply chain manager helped us validate the AI outputs against day-to-day needs. The IT lead ensured everything aligned with their internal systems.

We chose Gemini Pro for its high reasoning accuracy, supported by Azure and AWS services, GitHub for version control, Outlook for inbox integration, and OneDrive as the data hub. Its ability to interpret and reason through complex documents made it the right fit for the job.

To guide the AI agent’s behaviour, our developer created a tailored set of prompts and variables that clearly defined what to extract, what to ignore, and how to handle edge cases. A key part of the setup involved connecting each incoming purchase order to client-specific data stored in OneDrive. This made the system both flexible and future-proof. If contact details or client records need to be updated, the team can make those changes directly in OneDrive. The agent will use the updated information immediately, with no code changes or development work required.

The end result

The project wrapped in just three weeks, and Mi Integration’s supply chain team is already putting the solution to work in real-world conditions. The finished product transforms incoming purchase orders from PDF format into clean, structured CSV files offering a smoother, faster way to work.

Here’s what the new workflow looks like:

A buyer sends a purchase order via email
The AI agent scans the inbox every hour (during working hours), checking for new messages
For each new message, it matches the sender’s email address with client records and checks if there’s a purchase order attached (PDF or CSV)
If a match is found, the agent reads the file and cross-references the client’s record in OneDrive (if applicable)
It extracts detailed information from both the purchase order and the client record, then generates a new CSV with all fields mapped to predefined columns
The CSV is automatically saved to OneDrive using a consistent file name (including buyer name and order date)
The supply chain team accesses the file and continues operations as usual

What Mi Integration values most is the combination of accuracy, flexibility, and autonomy. The manual copy-pasting is gone. The AI agent now handles parsing and extraction automatically, which frees up the team to focus on higher-value tasks. And the results speak for themselves. When the supply chain team compared the AI-generated CSVs to their previous manual versions, not only were they accurate, they often caught subtle details that would have been easy to miss with the human eye. And while Mi Integration initially worried that every tweak would require outside help, their internal developer now has full visibility into the system and can make routine updates without relying on external support.

What’s next

With purchase orders now automated, Mi Integration is already thinking about what else agentic AI can tackle. Invoices, supplier forms, and other repetitive workflows are all on the radar. The foundation is already in place. Their systems are connected, their team understands how the solution works, and the architecture was designed to scale. Future development will focus on expanding functionality, not rebuilding from scratch.

This project proved what’s possible. And with the right setup, adding new use cases is a matter of when, not if.

Let’s build something together.

If your team is spending hours on manual, repetitive tasks, we can help you reclaim that time with automation that delivers real ROI. Ready to streamline a high-volume workflow? Get in touch.