Basketwise

WORKING VIDEO LINK BELOW

WORKING VIDEO LINK: https://www.loom.com/share/e862b9b5f91b4dc2af54c956af01b849

Inspiration

With the current landscape, things around us seem to be getting more and more expensive every year. We wanted to build something to help combat the issue of rising costs.

What it does

Basketwise is a proof of concept tool for finding the cheapest price for a given product. Simply provide the item(s) that you're looking for, your location via your zipcode, and the radius to search within. Basketwise then combs through all of the available online stores within range, to try and aggregate price data on the item(s) you requested. Once it finishes, you'll have a clear actionable list of items, ordered by price, and where you can buy it.

How we built it

Basketwise is written almost exclusively in Jac, as it simplifies database management, backend logic, front end, and more while helping save on context overhead for models trying to understand the codebase.

Challenges we ran into

Initially, we wanted to submit under the agentic track, and have Basketwise run with a set of agents who were responsible for combing through the web to find price information from different stores, another set of agents to iteratively discover undiscovered or niche grocery stores, and a deterministic backend for gathering and displaying the acquired data to the user. However, we ran into issues with assigning agents within the Jachammer environment. Beyond that, the price overhead of running those agents all but killed our initial ambitious goal, and forced us to reconsider how we would use AI.

Upon pivoting we still were faced with the issue of how to turn a sentence or request that's not consistent into a palpatable request for the AI. We used the byllm() call within Jac to handle this issue, it like dealt with the
We decided to pivot to backend

Accomplishments that we're proud of

We feel that proud that we got a working version of something thats full stack, that implements complex systems. All being accomplished with the new tool of Jachammer, which none of us have used before, showing our ability to learn and expand our horizons rapidly.

What we learned

One of the most important lessons we learned is that data is king. In our attempts to access content from various retailers, it was clear that asking agents to scrape was not sufficient. Most businesses locked access away behind APIs. Encountering those roadblocks also led to refining our agentic workflow. We learned to plan, prompt, review, and continue to loop this process till we developed the features we needed.

What's next for basketwise

We would like to set go back and set up things with our original structure. This would include returning to the original agentic approach as well as restructuring it to only use the api’s as a fall back. We would want to have our product work for all stores and businesses in the area, which would help be accomplished with our original web scraper esc. approach as well.

Another thing we would want to try out is building up our own api. With this we wouldnt need to rely on others api’s and we can have the most consistent and up to date data. With this we would also be able to include a more streamlined and usable interface.

As for expanding our product, we would look into fully developing into an app thats both supported on ios and android for broad use. We could also possibly integrate into an existing architecture such as the shopping section on google, which we believe would be a great home for our product.