Devlog #4
I added a goal feature that allows you to make goals and uses the AI to track the goals and see how well you are doing for the goals
devlog #3
I have added leagues into the app. I have different leagues based on how many focused hours you have done. I am hoping this will motivate more people into spending more time on this app and in turn being more productive in their life!
Ship
Devlog #2. FocusBuddy V1 is complete. Here is the overview
Tracker
Focus Buddy is a productivity tool that helps students start focus sessions and understand how focused they were during that time.
During a session, Focus Buddy classifies the user’s time into four states: focused, distracted, uncertain, and away. It does this by using webcam frames together with the task the user is working on. For example, if the task is “read a biology chapter,” the AI uses both the webcam frame and the task name/description to decide whether the user appears to be on task.
I also built a browser extension that can optionally give Focus Buddy more context from the active tab. For example, if the task is “write an essay,” the extension can share the active website title and URL. This helps the AI make a more accurate prediction because it can compare what the user said they wanted to do with what they are actually doing.
Session Debrief, Daily Unwind, and Weekly Unwind
For the session debrief, Focus Buddy saves the session data and then generates an AI debrief. This debrief explains what happened during the session and gives practical feedback for next time. It points to specific moments from the session instead of giving generic advice.
The daily unwind uses information from all the sessions recorded that day to give a general overview of what happened. During each session, the model records important changes in the user’s state. For example, if a user shifts from working on an essay to watching YouTube, the model can note that change. These state changes are then used as context for the daily unwind.
The daily unwind also compares the user’s focus level for that day with their recent average. For example, if the user’s average focus level over the past seven days was 80%, but they recorded 62% today, the model can give encouraging advice while still pointing out the main distractions or struggles from the day.
|
The weekly unwind works similarly, but it looks at a longer period of time. I also implemented a pattern document that stores general patterns about the user. These patterns can be affirmed or updated by the model after each session. The weekly unwind uses this information to give a recap of the week, including average focus levels, the user’s most focused days, and insights into their focus habits.
Plan Feature
The next feature I built is the plan feature. This allows the user to plan out their day before starting their work. They can pick tasks, schedule them for specific times, estimate how difficult each task will be, and estimate how long each task will take.
Based on this information, the AI can suggest the best times to work on certain tasks using patterns it has learned from the user. This step is completely optional, especially because this feature is still being optimized to provide better suggestions.
Started working on a productivity tool called focus buddy. I currently have a basic landing page with the ability to create tasks and a calendar for session review. When you go into a session the project takes information from the webcam and then it using the Gemini 3.5 flash model along with with the context of the task description and task name assigns one of 4 labels: focused, away, unfocused, and uncertain. Also added a analytics page where you can see your previous sessions + dark mode.
