Study Spot

Inspiration

We built Study Spot for people who struggle to stay engaged with their work, whether that’s because of anhedonia, ADHD, or just the everyday difficulty of staying focused. Everyone hits moments where their mind drifts, and we wanted to create something that gently supports people through those moments without being intrusive or distracting.

What it does

Study Spot is a desktop application that uses computer vision and adaptive math to notice when someone’s focus is slipping. As the user works, the Study Spot Agent watches for subtle aggregated signs of disengagement (facial micro-movements, keylogger inactivity) and uses a rolling buffer of snapshots of their screen to help the LLM contextualize what exactly the user was working on. When disengagement is detected, an AI assistant then looks at what’s happening and offers simple, helpful suggestions to get the user unstuck. Over time, Study Spot learns each person’s habits and becomes better at understanding what disengagement looks like for them. The result is a tool that feels personal, supportive, and genuinely helpful at keeping people on track.

How we built it

The application is split into 3 primary layers: the data processing layer, the user context layer, and the assistance layer.

At the data processing layer, we run two independent processes: one that collect metrics correlated with user engagement with a task at hand, and one that processes that data, and compares it to personalized metric thresholds based on user behaviour to determine when a user has become disengaged. This information is then used by the agent to decide when to intervene at appropriate times.

At the user context layer, we run another process that holds a rolling collection of images of the user's screen in order for the LLM to infer exactly what the user's task currently is, and what their specific goal or next step should be.

At the assistance layer, we pass in the user context and allow users to continue prompting the AI to gain further inspiration.

We built the UI using React.js. Our local backend process used Python and Flask for developing the REST API, handling all of our computer vision and data processing needs, data persistence for user preferences, and the rolling context window. Our application was packaged using electron.

Challenges we ran into

The integration between the computer vision models and the LLM-based AI assistant was an extremely challenging blend considering how different the nature of the two applications are.

Accomplishments that we're proud of

We're proud of the overall user experience we were able to provide, attempting to push for a more seamless and frictionless process of getting assistance, especially with those that struggle with executive function in terms of engagement. Specifically, we're proud that we're pioneering a transitioning from users prompting AI assistants to AI assistants prompting users.

What we learned

We learned how to program as a team and how important having a clear vision is for a product

What's next for Study Spot

Machine learning! We want Study Spot to feel personal to every user. We plan on integrating deep learning models in the future to make thresholds for disengagement even more accurate and personalized.