MindScope

MindScope passively detects cognitive overload from your work patterns and turns your task brain dump into a prioritized plan—what to do, schedule, delegate, or drop.

Comment

Inspiration

Knowledge workers burn out not from one hard day, but from the slow accumulation of fragmented attention, constant context switching, and never knowing when they've hit their limit. We wanted to build something that passively monitors cognitive load without adding another tool or task to the user’s workflow.

What it does

MindScope reads desktop activity via ActivityWatch (things like app switching, focus streaks, and idle patterns) and scores overload risk in real time against a personal baseline. When sustained overload is detected, it sends an email alert with the score and actionable recommendations.

A built in Task Copilot (powered by Claude) lets users brain-dump their to-do list and instantly generates:

  • an Effort–Impact Matrix
  • delegation suggestions
  • a time-blocked daily plan
  • a one-click .ics calendar export

How we built it

Backend: FastAPI + Python
Live ActivityWatch ingestion, a composite overload scoring model (fragmentation, focus instability, interruption load), scenario centroid matching, and SMTP email alerts.

Frontend: Next.js 14 (App Router)
Live dashboard with a trend chart, subsystem scores, and the Task Copilot interface.

AI: Anthropic Claude (claude-haiku-4-5)
Two-mode prompting for conversational replies and structured task-prioritization JSON output.

Baseline model:
2,400 synthetic activity windows used to generate a personal baseline profile for deviation scoring.

Challenges we ran into

  • Biggest challenge was ensuring that the data was accurately providing a proxy for feeling overwhelmed - may need more fine tuning to ensure baseline profiles/behavioral groups are accurate. This came up many times with false positives and many emails that were not accurate reflections of overloaded score.

Accomplishments that we're proud of

  • We initially thought about just having a copilot that could help organize, prioritize and delegate tasks, but happy that we found a real time data source like 'Activity watch' that really elevates this product, and allows a trigger for emails / help.

What we learned

  • Personal baselines matter far more than universal thresholds

What's next for MindScope

  • Biggest thing to increase the robustness of this model is to pair it with our signals (things like wearable integrations (HRV / Oura Ring) to combine physiological and behavioral signals)

Built With

Share this project:

Updates