What I Built
I built Daywell, a personal planner for my friend who has trouble keeping a track of their daily activities and ends up over scheduling. It combines tasks, reminders, a daily schedule, and productivity insights to help make busy days more manageable.
Daywell calculates planned and available time, flags overloaded schedules, and suggests tasks to move. Over time, it tracks what gets completed or postponed, so its monthly analytics can reveal patterns—like which days or hours tend to be most productive.
Demo
There isn’t a hosted demo yet. You can run the app locally by following the instructions in the README.
Code
View the Daywell GitHub repository: https://github.com/pathrikarsatwit/AI-Reminders
How I Built It
Daywell is a responsive React, TypeScript, and Vite web app. Tasks and productivity history are stored in the browser. Scheduling and analytics are calculated by deterministic application code—not by an LLM.
For AI insights, I built a provider boundary around Google’s open-weight Gemma, served locally with Ollama. When someone requests insights, Daywell sends aggregated statistics and supporting evidence to their local inference server; it does not send raw task names or notes. This keeps the AI optional and leaves room for other local inference setups in the future.
I also used GitHub Copilot during development to help build the app incrementally and refine its features.
Why Does Open Innovation Matter?
Productivity data is personal. Using an open-weight model with local inference makes it possible to get AI-generated observations without making a proprietary cloud AI API the core of the app. The app can keep task history in the browser and share only aggregate data with a local model when the user asks for insights.
Open tools and models also make the AI layer easier to inspect, adapt, and extend as local inference options evolve.
Prize Categories
• Best Use of GitHub Copilot
Top comments (1)
This is actually a really interesting approach to daily planning. I especially liked the idea of detecting when someone's schedule is overloaded instead of just generating a timetable that looks perfect but is almost impossible to follow.
I've been working on a similar problem from a slightly different angle. I built an app called Personal Planner, where the main focus is comparing how much time you planned for a task with how much time you actually spent on it, and then using daily and weekly reviews to understand those differences.
It recently crossed around 200 downloads, which has been a pretty exciting milestone for my first app.
What I find interesting is that Daywell tries to make the plan realistic beforehand, while my approach focuses more on learning from what actually happened afterward. I feel like both are important parts of the same problem.
One thing I'm curious about: does Daywell eventually adjust future time estimates based on how long previous tasks actually took?
Really nice work on this. Would love to see how you develop it further!
If you want to checkout the app: github.com/Karthikshabari/Personal...
Few images
