AutoSupply

Inspiration

The inspiration for AutoSupply was taken from one of our colleagues working in one of the biggest conglomerates of India. The company has very set standard operating procedures when it comes to inventory management, reordering and vendor management. After hearing about what the company does, we got the hint of why not automate this entire process via an AI Agent built right into the browser.

What it does

AutoSupply is an AI agent built right into your browser, which can go from your inventory right to the checkout page across all orders. It works by identifying the needs of the company on the basis of its inventory, which includes the quantity which needs to be reordered and the vendor that will fulfill this reorder. Then, it automatically places these orders depending upon the types of orders they are. If the order is placed by sending purchase orders via mail, the agent will automatically draft purchase orders, take confirmation from the user and send them to the right vendor. If the order is placed through online websites, the agent will automatically browse the website, find the product, change the quantity, add to cart, and proceed to checkout.

How does it involve Gemini

AutoSupply as an agent sends backend calls to Gemini lightweight models. These API calls generally involve drafting emails to be sent as purchase orders, analyzing user inventory via a JSON, reasoning what is to be ordered with what quantity and other related agentic behavior. Any questions that the user asks to this agent are also responded to by Gemini, by taking the user inventory in question as the only form of context available.

No payment or financial decision is taken by the agent and that is completely left to the user.

How we built it

The agent was built completely in Javascript. For the backend which listens to all AI calls, drafts mails, purchase orders and sends the mail itself, we created a Node.JS server. The frontend of the agent is created using Vanila HTML + Tailwind CSS and its backend is also completely handled in Javascript.

Challenges we ran into

One of the major challenges that we ran into while building this agent was ensuring uniform performance across different websites for online orders. There are a lot of websites which represent industry standards across various industries. To achieve the same performance throughout websites where we have to change quantity, add to cart, go to the checkout page was very difficult as different websites adopt different UX for the same.

Accomplishments that we're proud of

We are proud of the system that we have built, as this is not automation on just one side, but it caters to much more than that. Not only are we limited to online orders, but we do email purchase orders as well. The agent has a very sleek and simple-to-understand frontend, which has taken a lot of its inspiration from Claude Code and CoPilot Agent on VS Code.

What's next for AutoSupply

For AutoSupply, the next step is ERP Integration. Since the agent right now can only scan inventories that are in google sheets, next step for us is to integrate it in Industry Standard ERP Softwares. There are multiple ways by which we can go about this, the easiest one of them is to integrate into softwares which let you export your inventory as an Excel file or a CSV, and then upload that to Google Sheets to integrate it with AutoSupply, but this takes control out of the ERP and is not preferred. Once AutoSupply has scaled enough, we plan on partnering and integrating with state-of-the-art ERP softwares.