WanderWithin

• 

  Model takes the input of how a person is feeling at the moment, and tries to provide help.

• 

  Focusing on the details of the issue

• 

  App provides a response in the Web UI

• 

  Looking at the bigger picture

• 

  Entry point

• 

  Reflecting current thoughts

Inspiration

The inspiration for WanderWithin came from the desire to help individuals explore their thoughts more deeply. We wanted to create a tool that guides users to view their experiences from multiple perspectives—much like Alice’s journey in Wonderland, where her perception of size and scale constantly shifts. This approach encourages users to reflect on their challenges from both detailed and broader viewpoints, promoting a more balanced perspective.

What it does

WanderWithin is a digital journaling tool that uses "shrink" and "grow" metaphors to help users reflect on their emotions, thoughts, and actions. It guides users through a series of prompts to analyze their experiences in fine detail ("shrink") and then encourages them to look at the bigger picture ("grow"). The platform provides insights and suggestions to help users understand their emotions better and find actionable paths forward.

How we built it

We developed the UI using Angular to create a smooth, engaging experience for users. For the core reflection features, we used a language model (LLM) built using Hugging Face Transformers. The model provided tailored insights based on user input. We initially attempted to host our model on AWS SageMaker, but we faced challenges due to the model size exceeding the free-tier storage limit.

Challenges we ran into

The main challenge we faced was with hosting the LLM. The model's size was around 5 GB, which exceeded the storage capacity available for free on AWS SageMaker. Despite compressing and optimizing the model, we found that the required GPU resources and storage limits were beyond the available free-tier resources. This prevented us from successfully deploying the model as an endpoint on AWS.

Accomplishments that we're proud of

We’re proud of building an interactive and intuitive UI using Angular that makes the journaling experience engaging. We also successfully fine-tuned an LLM with Hugging Face Transformers to generate meaningful reflections, making the tool highly personalized. Despite the hosting challenges, we learned a lot about cloud deployment options and managed to fully integrate the core components locally.

What we learned

We learned the intricacies of working with LLMs and deploying them using cloud platforms. Hosting models larger than a certain size often requires careful resource planning or optimization, especially if cost is a constraint. We also learned the importance of adapting to different scenarios—when AWS didn't work for free, we explored alternatives like Google Colab and other cloud services to keep moving forward.

What's next for WanderWithin

Next, we plan to:

Optimize the LLM to reduce its size, possibly through quantization or using a smaller variant. Explore other cloud providers that provide free GPU resources. Continue enhancing the UI to make it more immersive by adding animations and themes inspired by Alice in Wonderland. Deploy the project using more efficient infrastructure to make it publicly accessible for users who want to reflect on their thoughts in an insightful way.

Built With

• python
• transformers