Second Thoughts

Gameplay
Case

Inspiration

This app is inspired to oppose google pay, stripe, klarna, plaid - and all apps supposedly for your benefit in making transactions convenient and pleasant. In reality, where trillions are spent by high-budget multinational corporations analysing your activity patterns to promote things you don't actually need - convenient transactions cause regret and reckless spending.

What it does

Our web-app 'Second Thoughts' uses an extension to intercept transactions and have you answer a series of reflective questions, forcing your decisions from reactive - to proactive, and promoting good financial spending.

How we built it

We used a three js with React for the frontend to create a interrogation room scene. An extension using pure js was built to scrape any site that allows purchases which include any words that have "Buy", "Buy Now", "Add to basket" etc to then redirect to the main web app. It also sends a HTTP request to the site and polls the site every 2 seconds to see if there is a verdict, once reached it will replace the fake buttons back and th euser can continue buying

Challenges we ran into

Engineering the prompt for the AI backend so that it wasn't strict but was also still in character. This required multiple tests and iterations. The extension had a couple of issues with context, it didn't know whether a number on the screen was related to the item or just a random number so after adding scopes that was fixed.

Accomplishments that we're proud of

We're proud of the idea: one-click payment took all the friction out of spending, so we built a fun way to stop it. The frontend is a fully 3D interrogation room in dithered pixel art, with a swinging bulb, a tape recorder and a case file you can lean in to read. It plugs straight into real shopping. Our extension catches the pay button on real shops and sends you to the interrogation room, and the button only comes back if you convince the detective. The final product is live and playable end to end: click Buy now on a real shop, plead your case, and see whether you're allowed to pay!

What we learned

We learned about limiting the scope of a project, as well as the challenges that come from integrating an LLM into an app - the output is unreliable, and we struggled with getting satisfactory answers. Some techniques that helped were constraining the number of responses and providing examples of chats with the output we wanted. Eventually, we got a satisfactory result, but there's definitely more we could do to improve it further.

What's next for Second Thoughts

As a proof of concept, I think it would be fun to spin out this app into a few different directions. One would be a more serious, clean project that appeals to daily use, giving logical pros and cons behind a purchase. Another would be further gamifying the app - a concept like this could work well as a standalone game.