Wit App ID: 328662241461519
Will has long talked about a management system for the home, a seamless way to keep on top of building maintenance and create a record of work done. Together, we came up with a chat-based solution for keeping track of work done, work to do, and even recommendations for your home. We think there's a lot of potential in scheduling, locating contractors, and keeping a historical building record!
What it does
When you first start using Wkend, you'll be asked to give your home a name and describe it a little bit. Using Wit.ai, we were able to parse those attributes and begin creating a record for the home. From there, you can tell Wkend tasks you need to do regularly, or reference when they were last done.
How we built it
Our first decision was to keep this web based for ease of accessibility. We turned to Next.js and ANT for our UI, and FastAPI for our server. After doing a bit of training in Wit.ai, we were able to set up simple endpoints for handling text or speech and returning responses to the user. One neat integration we were able to take advantage of is the text-to-speech browser api now widely available, which coupled with Wit.ai allowed for a full speech interface. For authorization we integrated Auth0, and in order to save the user's data we set up Hasura with Postgresql. This combination allowed for a slick web application and a fully featured backend.
Challenges we ran into
We ran into so many challenges! Recording in different browser contexts, sending speech snippets to FastAPI, trying to launch the Next.js app on AWS lambda... the list goes on. We had high hopes for completing a fully fleshed out application, but at the end of the day we were both excited to take our first stab at chat/voice enabled app.
Perhaps one of the more surprising challenges has been comprehending the types of interactions a chat-based interface entails. How do we make it clear to a user the actions available? How do we handle different chains of thought that end up at the same result?
Accomplishments that we're proud of
A few of our accomplishments were launching FastAPI on lambda, getting both sides of the voice interaction to work, and overall architecting a multifaceted application. The combination of libraries we arrived at seems very versatile, and we're excited to push Wkend further.
What we learned
Don't try too many new technologies at once! Oh my goodness, we overwhelmed ourselves and were not able to get to a super cohesive application. That's not to say we have any immediate regrets, learning to use many of these new tools such as Next.js and Wit.ai has made it all worthwhile.
What's next for Wkend
We're going to continue developing Wkend until it can be properly deployed, tested out in the real world, and then shared with a few friends we know who are interested in using it. There are still a few key features we would like to get to, such as searching for contractors or providing recommended cost-saving tasks. We'd also like to provide Alexa or Google Home integration so interactions could be more passive--the goal here really is simplifying staying on top of house work and keeping a record of it.
Thanks for the opportunity to be part of this Hackathon!