Habit Step

Problem Statement 🤔

Many people wish they had productive habits but always struggle to create them. Routines that are meant to help create habits are often too intense, not focused enough, or contain steps that aren’t possible due to either disability or lack of resources. Ultimately, plans lack personalization and specificity. Users want access to routines that create and reinforce habits so that the process of habit-creation is optimized for success.

Inspiration 💡

Strong passion for AI and its integration into everyday life.
Desire to provide users with a highly curated plan users can use to achieve their goals.
Belief in the potential of AI to aid humans in achieving goals and accomplishing new things.

What it does 🎯

Our solution, Habit Step aims to solve the problem by creating habits tailored to both your needs and your restrictions. Habit plans are created using a detailed questionnaire and the help of AI to guarantee maximal personalization.

How we built it 🛠️

Habit Step was built with simplicity in mind. It implements a minimalistic UI to keep distractions away from the user experience
PostgreSQL hosted by Neon was used as the backend data storage mechanism due to its relational model. Postgres was especially helpful when storing multiple choice questions since we could use the JSONB column type to store JSON directly
OpenAI’s gpt3.5-turbo model via the chat completion API was a natural choice due to our application’s need for large contexts and significant completion sizes. OpenAI helps us generate specific questions according to the user’s habit needs as well as create the habit plans themselves.
We make use of a zero-shot text classification model hosted by HuggingFace to sort a user’s habit input into one of a few categories so that we can give users a pre-generated question set that is targeted at that category. We generate the remainder of the questionnaire with OpenAI so that we maintain specificity while reducing latency by leveraging our own question set
Nextjs 13 and React Server Components (RSC) were a big part of our frontend/backend development experience as they let us co-locate data fetching and render templates to speed up development time

Challenges we ran into 🧩

Formatting responses from GPT3.5 into a consistent JSON structure.
Determining the best balance between high specificity questions and low latency for the end user.
Ultimately, we believe the balance we struck is the optimal middle ground between both

Accomplishments that we're proud of 🌟

Developed a sleek and minimalist user interface
Integrated both ChatGPT and Facebook’s zero-shot text classification model into the app, providing powerful analysis and responses driven by user data.
Utilized server side components with Nextjs 13 to boost app performance and drastically decrease page load times
Created a user funnel to streamline the habit creation process, making the app simple and easy to use

What we learned 🧠

Gained better understanding of how an application can make use of HuggingFace’s hosted APIs to -leverage a variety of models to complete complex tasks
Learned how to make ChatGPT and similar NLP models return predictable and formatted responses that can be used for further analysis.
Learned more about the latest release of Nextjs and gained more experience with the beta version of the app directory
Realized the importance of proper time management and task delegation.

What's next for Habit Step 🚀

Include additional categories and subcategories and build out their associated question sets so users have low-latency access to rich question sets that are specific to their desired habit
Optimize requests to OpenAI such that the responses are streamed to the client one-by-one as opposed to the single big-batch request we make at the moment

Why Habit Step? 🥁

Habit Step should win multiple awards this hackathon, including Best AI Hack, Best Themed Hack and Best UI/UX Hack. We believe that our project stands out because we’ve successfully integrated two separate AI technologies that work together to create a seamless and highly curated user experience. This combined with the simplistic user interface makes our application easy to use for everyone. Here’s how we’ve met each challenge:

Best AI Hack: Habit Step leverages two forms of AI to achieve its tailored questionnaire and habit-generation. It makes use of zero-shot text classification and generative text models to ensure every user experience is unique.
Best Themed Hack: Using the latest and greatest of AI, we’re able to dramatically increase a user’s potential for achieving a higher quality of life. The end result of Habit Step is a user as empowered as they can possibly be to change their lives for the better.
Best UI/UX Hack: Habit Step maintains a minimalistic UI with a simplistic UX that gets users into our primary user flow without distraction. No matter a user’s desired habit, Habit Step gets them right into a tailored questionnaire to best create their habit routine. This level of flexibility means that our user experience is as accommodating as it can possibly be.