Roastii: Roast My Receipts!

Roast your shopping habits with Roastii, an adorable AI companion who cuts deep to protect your wallet. Roastii stops you from making impulsive purchases on Amazon.ca, before you click "Add to cart".

Comment

See the instructions on our GitHub repo to try our project!

Inspiration

As tech enthusiasts, we avidly shop for the latest and most exciting new tech. As graduate students, we are broke. As a result, we often have to talk each other down on impulsive purchases that seem perfect in the moment, but which we will regret in the few days after purchasing. We built Roastii to help with these habits, while also having a little bit of fun with some creative roasting.

What it does

Roastii is an adorable AI companion wrapped into a Chrome Extension. Upon installing, she asks for your shopping lore, prompting you to complete a few questions so that she can build a profile on your financial status and needs. An example question would be, "Which of these brands/categories makes your wallet cry?" After completing the profile, you can feel free to start shopping on Amazon.ca. When you add an item to your cart, Roastii will pop up and roast your decisions! Past roasts will be saved, allowing for dynamic improvement of roasting as she monitors your habits.

How we built it

We experimented with a few AI agents during the development process, and ultimately had to correct significant bugs and logical misunderstandings ourselves throughout the development process. Tools used include GitHub Copilot, Claude Code, OpenCode, Groq and OpenRouter. Models include Qwen 3.5, Claude Sonnet, GPT-OSS and GPT-5.4 and a handful of others.

The extension was built in Typescript, with HTML and CSS page content, then transpiled into JavaScript before being packaged as a .zip file for distribution on GitHub. We use OpenAI API Library, means you can basically use any LLM provider that is OpenAI Compatible.

Challenges we ran into

AI hosting was a challenge. We have an M4 Mac Mini that we originally wanted to use as the backend because we thought running on our own hardware would be cheaper and easier. However, we encountered several performance issues with this hardware. We instead decided to use Groq, as their model inference is so fast that we don't even need to use the Streaming feature! Perfect for scenarios requires instant response. Slighty over 70% of the code were genereated using AI tools, but they were buggy and we fix most of the bugs manually.

Accomplishments that we're proud of

In the end, we created a reliable, high performance application. Roastii adjusts her attitude according to our personal profile and currently viewing item, which will totally crush our shopping impulses. We really like the look and feel of the app as well as its simplicity.

What we learned

Both of us experimented with new tools that we've never used before. James: While I'm a longtime software developer and eager tech enthusiast, I've been out of the loop of the AI trend for a few years which was enough time for everything to be way over my head. GenAI Genesis was a valuable opportunity to practice and sharpen my skills in the field and catch up on everything I've missed! Since my PhD research will employ LLMs, this experience was quite valuable for me. Benz: I have experience with collobrating with agentic coding tools and I am really glad I could put this to use in another coding projects! For me, GenAI Genesis gave me the opportunity to code a unique project under the challenging constraints of a hackathon. Over this weekend, I learned a new programming language and sharpened my rapid prototyping skills.

What's next for Roastii: Roast My Receipts!

We want to integrate image capture and OCR, so that Roastii can become universal and roast purchases from anywhere, just by taking a picture with your phone! We also wish to integrate with other websites, like Best Buy, Aliexpress, Temu, etc.

Built With

Share this project:

Updates