Mumble Sort: AI task organizer

A simple, yet powerful, task management app to organize thoughts & tasks with AI. The perfect tool for busy founders and makers to save time and ensure and idea is never lost.

Comment

Inspiration

I've always been a fan of productivity tools, but I found many existing solutions to be overly complex or lacking features, especially for solo developers or makers. I wanted to create an app that could understand my unique workflow and adapt to my needs: writing tasks quickly, managing them in a simple view, and helping me organize them. AI integration seemed like the ideal solution.

What it does

MumbleSort is an AI-powered task organizer designed to streamline your workflow and boost productivity with features like:

  • Simple Kanban style: manage all your tasks with a simple TODO/Waiting/Done flow
  • Voice input: just say what's on your head and let the AI summarize it and organize it into tasks
  • Ask and brainstorm with the AI: ask questions related to your tasks and get answers right away (e.g., "What should I focus on today?")

How we built it

I used my own starter-kit (ShipFlutter) to bootstrap the project. This includes:

  • Flutter
  • Firebase integrations (auth, firestore, etc.)
  • GenKit integration
  • RevenueCat SDK

Afterwards, I modified the different modules and added the required features.

Challenges we ran into

The biggest hurdle I overcame was ensuring that the AI model could accurately interpret and understand user input, especially when dealing with ambiguous or complex queries. To address this, I used GenKit with the Gemini Flash model together with Firebase Functions. This setup creates a vector dataset whenever a task is created or updated. This data is then used to feed the Retrieval-Augmented Generation (RAG) AI model.

Accomplishments that we're proud of

I really like the result of the UI. I find it simple and easy to use. But what I'm most proud of is the backend using GenKit and RAG to handle user input. It opens the door to a lot of possibilities since I don't have to manually code each available action. I provide the AI with "tools" and let it decide what to do.

What we learned

  • GenKit integration
  • Retrieval-Augmented Generation (RAG) with Firestore

What's next?

A lot of improvements! Due to some issues and other commitments, I had to speed up development. First, I want to improve the AI model and then focus on optimizing the payments flow.

Built With

Share this project:

Updates