TimeSense
2026 · Solo build · Active
Mobile-first, context-aware personal time assistant. Learns your routines, commute, meals, sleep, and calendar to tell you what to do next — without making you maintain another productivity system.
Demo
Highlights
- Native on both platforms — SwiftUI and Jetpack Compose, not a web wrapper. The web companion handles account setup and billing, not daily use.
- Recommends a single best next action from calendar, location, time of day, priority, and observed energy, with a confidence score attached.
- Every suggestion is explainable: the app shows which signals drove it and which alternatives it considered and rejected.
- Learns from accepted and rejected suggestions, so recommendations sharpen as history accumulates.
- The LLM layer is provider-agnostic, so the reasoning backend can be swapped without touching product code.
Background
TimeSense started from a simple observation: most productivity tools add work. They ask you to maintain lists, tag tasks, and groom backlogs — and the maintenance quietly becomes another job.
TimeSense inverts that. It reads the signals already around you — calendar, location, time of day, activity, sleep — and answers one question: what should I do right now? You can accept or reject its suggestion, and that feedback is the only input it really needs.
The explainability work matters most to me. A recommendation you cannot interrogate is one you will not trust, so every suggestion opens into the signals behind it and the alternatives that were weighed. That connects directly to my research on trustworthy AI: a system that hedges or hallucinates its reasoning is worse than one that says less but means it.
Stack
- Swift / SwiftUI (iOS)
- Kotlin / Jetpack Compose (Android)
- FastAPI
- PostgreSQL
- Redis + Celery
- Firebase Auth
- Stripe · StoreKit · Play Billing
- Provider-agnostic LLM layer