This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend.
🧠 Unmess — Turn the chaos in your head into a plan
I built Unmess, a local-first AI productivity assistant for a friend who often has too many things going on at once and struggles with turning all of those thoughts into a clean to-do list.
The problem is simple:
People don't think in task lists.
We think in messy paragraphs, fragments, reminders, worries, deadlines and random thoughts.
So instead of asking someone:
"What tasks do you want to create?"
I wanted Unmess to ask:
"What's on your mind?"
💡 What I Built
.
Unmess takes a messy brain dump and turns it into structured, actionable tasks.
For example, someone might write:
I need to finish my homework tomorrow, submit my exam form by midnight, study for my test on October 6th, send flowers to my friend, and buy groceries today.
A traditional productivity app expects you to manually turn that into five separate tasks.
Unmess does that part for you.
It can:
- 🧠 Extract individual tasks
- 📅 Detect natural-language deadlines
- ⚡ Assign priorities
- 🔗 Identify dependencies between tasks
- 💾 Persist tasks for later
- 🤖 Use a local open-weight AI model
Input
I need to finish my homework tomorrow,
submit my exam form by midnight,
study for my test on October 6th,
send flowers to my friend,
and buy groceries today.
### Output
text
🔴 Homework Assignment
Deadline: Tomorrow
Priority: Critical
⚠️ Submit Exam Form
Deadline: Midnight
Priority: High
📚 Study for Test
Deadline: October 6
Priority: High
💐 Send Flowers
Priority: Low
🛒 Buy Groceries
Deadline: Today
Priority: Low
The core workflow looks like this:
Brain Dump
↓
AI understands the input
↓
Tasks extracted
↓
Deadlines detected
↓
Priorities assigned
↓
Dependencies identified
↓
Tasks stored
GitHub Repository:
https://github.com/gaurimshra/unmesss
https://github.com/gaurimshra/unmess
🛠️ How I Built It
I wanted the architecture to stay simple enough to understand while still demonstrating an actual AI workflow.
Frontend
Next.js
React
TypeScript
Tailwind CSS
**Backend
**Python
FastAPI
Pydantic
AI
Gemma 3 1B
Ollama
Local inference
Storage
SQLite
SQLAlchemy
The architecture is roughly:
👤 User
│
▼
┌─────────────────┐
│ Next.js Frontend│
└────────┬────────┘
│ HTTP
▼
┌─────────────────┐
│ FastAPI Backend │
└────────┬────────┘
▼
┌───────────┐
│ AI Agent │
└─────┬─────┘
│
▼
┌─────────────────┐
│ Ollama + Gemma │
└────────┬────────┘
│
▼
Task / Planner Tools
│
▼
┌─────────────────┐
│ SQLite Database │
└─────────────────┘
]
🔓 Why Does Open Innovation Matter?
This is one of the parts of the project I care about most.
A productivity assistant can potentially see extremely personal information:
unfinished work
deadlines
personal plans
private thoughts
worries
responsibilities
random ideas
So I didn't want the core concept to depend entirely on sending those thoughts to a closed third-party AI API.
Unmess uses Gemma 3 1B through Ollama, allowing the AI component to run locally.
That gives the project a different privacy model:
Your Thoughts
↓
Local Application
↓
Local AI Model
↓
Local Database
The bigger idea is:
Your thoughts shouldn't have to leave your machine just to become organized.
Open-weight models make it possible to experiment with AI systems where the model and inference environment can be much closer to the user.
That's especially interesting for applications that deal with personal data.
👭 Build for a Friend
I built Unmess around a very real problem that I see in people who have too much going on at once.
Instead of asking someone to maintain another perfectly organized productivity system, I wanted to make something that works with the way people actually think.
You don't always think:
"Create task → set priority → set deadline."
You think:
"Oh crap, I also have to submit that form tomorrow and I haven't even started."
That's the kind of input I wanted Unmess to understand.
**
🚧 What's Next?
**
Unmess is still a work in progress.
Some of the things I want to build next are:
✅ Task completion
✏️ Task editing
🗑️ Task deletion
🧠 Persistent AI memory
💭 Random-thought organization
🏷️ Automatic categorization
🔔 Deadline reminders
📅 Calendar integration
🔗 Better dependency planning
🤖 More autonomous task planning
Eventually, I want Unmess to become more than a task extractor.
I want it to become an AI layer between your thoughts and your actions.
**🌱 The Vision
**
Unmess is built around one simple idea:
Your productivity system shouldn't require you to already be productive.
Real thoughts are messy.
Deadlines overlap.
Responsibilities are scattered across conversations, notes, memory and random thoughts.
You shouldn't have to clean up your brain before you can ask for help.
Messy Thoughts
↓
Understanding
↓
Tasks
↓
Priorities
↓
Plan
↓
Action
Just dump it. Unmess it.
**
🏆 Prize Category
**
Best Use of Gemma
Unmess uses Gemma 3 1B through Ollama as its local AI model for understanding and structuring brain dumps into actionable tasks.
The project uses an open-weight model as part of its core workflow rather than treating AI as an external black-box API.
** Built By**
_Gauri Mishra
_
Computer Science & Engineering student exploring:
AI Agents • Generative AI • Local AI • Developer Tools • Intelligent Productivity Systems
**
**
I'd gen
uinely love feedback on Unmess.
Would you use something like this?
What would you want an AI productivity assistant to remember?
And most importantly:
Would you trust a local-first AI more with your personal thoughts?
Thanks for reading ❤️
Top comments (0)