FlowBud
I built a browser extension (using a combination of Lovable and Replit) called FlowBud to help me maximise my daily output while minimising time spent on each task. Here are the specs:
FlowBud: Do more daily, focus for longer
Problem: Using the same devices for work and leisure means during work hours we catch ourselves opening YouTube or answering our mom on Signal. Mindless distractions lower our output/prolong how much time we take to complete tasks.
Solution: FlowBud is a gamified approach to 1. Help us to stay on track on the daily, and 2. prime us to become aware of our distractions and 3. extend our attention span over time.
How: A browser extension that detects when you get distracted and prompts you to
- commit to an extra 5/10 minutes or remaining task time,
- switch to a task with a low cognitive load or
- take a conscious break
Distraction identification: Initially, the model learns by asking the user whether they got distracted when certain behavioural benchmarks were met such as “looked away from screen for 25 seconds”, “Fidgeting increased by 20%”, “Reddit was opened”,.... Not only does the model learn with every answer it receives on pre-determined behaviors (Got distracted? Yes/No), it also picks up more behavioural patterns that inform distraction detection in future (e.g. if user answers “Yes”, other behavioural markers leading up to that distraction are being saved e.g. movement of the mouth).
Limitation: Detects false positives (identified but no distraction) but not true negatives (distraction but not identified)
Gamification:
1. Compete with yourself
2. Compete with averages of all users
3. Beat your all-time longest interval score
4. Beat last weeks stats:
- Set a focus interval goal (e.g. 25-45 minutes Pomodoro technique) and beat last weeks amount of intervals
- Beat last weeks focused minutes score
- Beat last weeks count of distractions
Audience: 1. People whose work largely takes place on the computer
- Especially those working remotely
- Especially those with ADD
- Especially those from younger generations (that struggle with shorter attention spans on average)
2. Companies/organizations with a largely remote team (FlowBud shall never be a workforce surveillance tool and the architecture ensures that)
Privacy: FlowBud is designed so that all sensitive data remains on the user’s device. The user governs their data, and the system is technically constrained to prevent access to raw webcam or behavioral data.
Information Flows:
1. User -> On-Device FlowBud System
What flows: Webcam signals (e.g. gaze, movement,..), browser context, user responses to prompts
Governance: Fully user-governed, all processing happens locally
Privacy needs:
- Input privacy: Required (highly sensitive data)
- Verification: Not required (data is not shared)
- How: On-device processing, no raw data storage or transmission
2. On-Device System -> User
What flows: Distraction notifications and focus prompts
Governance: Platform-governed logic, with the user as final authority
Privacy needs:
- Output privacy: Required (feedback is based only on the user’s own data)
- How: Local decision-making; no cross-user comparisons
3. On-Device Model -> Federated Learning System
What flows: Small, abstract model updates (no raw data or features)
Governance: Protocol-governed; updates are constrained
Privacy needs:
- Input privacy: Required
- Input verification: Required
- Output privacy: Required How: Federated learning, secure aggregation, differential privacy
4. Aggregated Model -> Users
What flows: Improved baseline model parameters.
Governance: Platform-governed.
Privacy needs:
- Output privacy: Required
- How: Differential privacy and aggregation so that no individual influence is exposed
