DEV Community: TanzeelAhmad

# Eatz: Know What's in Your Food, Without Sending It to the Cloud

TanzeelAhmad — Sun, 18 Oct 2026 18:06:48 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.

Disclaimer: Halal Scanner is still under active development. Features, ingredient data, and classification accuracy may change as development continues. The app is not a substitute for authoritative halal certification or guidance from qualified authorities.

What I Built

Halal Scanner is a mobile application designed to help Muslim consumers check ingredient labels on packaged food.

Reading ingredient lists isn't always straightforward. Ingredients can have unfamiliar names, manufacturers may use different terms for the same substance, and some ingredients require further verification before their halal status can be determined.

The idea behind Halal Scanner is simple: take a picture of an ingredient label, extract the text, and check the recognized ingredients against a local database.

The app categorizes results as Halal, Haram, or Uncertain. Instead of automatically treating unknown ingredients as permissible, it flags them for further verification.

The potential audience includes Muslim consumers who regularly shop for packaged food and halal-conscious retailers looking for a convenient ingredient-checking tool.

For the side-hustle angle, a potential business model could offer a free consumer app with optional premium features, such as an expanded ingredient database and personalized product lists. This is a proposed direction, not an existing paid service.

Demo

Project repository: Halal Scanner on GitHub

A demo video or installation link can be added here when available.

Code

Explore the source code and project implementation:

github.com/CodewithTanzeel/HF-Dev2-challenge

The application connects image capture, on-device text recognition, ingredient matching, classification, and local scan history into one mobile workflow.

How I Built It

I built Halal Scanner using React Native and Expo.

The scanning workflow consists of four main stages:

Capture: The user takes a picture of an ingredient label.
Recognize: On-device text recognition extracts the printed text using Google ML Kit.
Classify: A local rules engine matches recognized ingredients and their synonyms against the available ingredient data.
Save: Scan results can be stored locally for later reference.

Rather than relying on a generative model to make religious judgments, the current classification approach uses explicit rules. When an ingredient cannot be confidently matched, the app can return an uncertain result.

This approach makes the decision process easier to inspect and provides a clear way to handle missing information.

The project is still evolving, and the ingredient database, matching logic, and overall user experience will need continued testing and refinement.

Why Does Open Innovation Matter?

A food-label scanner shouldn't necessarily need to send every photograph to a remote service just to recognize its text.

On-device processing can reduce dependence on network connectivity, avoid per-request cloud inference costs, and help keep ingredient-label images on the device during text recognition.

Open tooling also makes it easier for developers to inspect the implementation, improve ingredient matching, expand the database, and contribute to the project.

There is an important limitation, though: a local database is only as reliable as its sources, coverage, and maintenance. Recognizing an ingredient doesn't automatically establish its halal status, and an uncertain result should encourage further verification rather than provide false confidence.

My goal is to make ingredient checking more accessible while keeping the technical decisions understandable and open to improvement.

My Agent Session

Optional: Add your DevRelay agent-session link here if you have one.

Prize Categories

Add the eligible partner categories whose technologies your project actually uses, based on the challenge requirements.

# I Built a Local AI Pet That Yells at You to Go Outside — No Cloud, No API Keys

TanzeelAhmad — Mon, 12 Oct 2026 05:26:51 +0000

Hacktoberfest Week 1 • Touch Grass Challenge

This weekend I submitted Miso — a memory-efficient desktop companion that runs a real open-weight LLM locally and literally tells you to touch grass when your focus timer hits zero.

GitHub logo CodewithTanzeel / touch-grasschallenge-miso
Miso — Touch Grass Desktop Companion
🤖 Built with open-source AI at its core.

What is Miso?
No cloud. Locally optimized. Memory efficient. Secure.

Miso is a cute, interactive desktop pet that sits in the corner of your screen. It works with you, not against you. The pet tracks how long you have been focused, and when your timer runs out, it gets angry (in a cute way) and literally tells you to go outside and touch grass using an open-weight language model running entirely on your laptop.

This is a Progressive Desktop Widget — fully offline, open-source, local-only, and designed specifically for low-end hardware (i7-6600U / 8GB RAM). Zero cloud APIs. Zero subscriptions. Zero data leaving your computer.

Why Open Source AI Matters Here
This project was built specifically for people with older, lower-spec hardware (like an i7-6600U with 8GB RAM). Rather than relying on expensive cloud APIs or…

View on GitHub
What it does
Miso is a cute, interactive desktop pet (Pikachu-style) that tracks how long you've been focused. When the timer runs out, it gets angry — in its cute way — and uses a local LLM (SmolLM2 / Qwen2.5) to generate a unique scolding via TTS. No cloud APIs. Zero subscriptions. Zero data leaving your machine.

Features:

Hover interaction: dance and emotions
Drag-to-position widget
Idle tracking: calms down if you actually leave
Universal customization (MISO_CHARACTER=my_pet)
Character hub: drop GIFs + config.json to build your own
The real work this week (Hacktoberfest fixes)
The PRs I submitted cover:

Deprecated param fixed
Download script repaired
Memory-efficient local AI configured
Memory file added (local/secure tracking)
.gitignore, docs, tests, TTS fallback, install errors cleaned up
User guide + emphasis on local/secure/memory-efficient
Qt6 / MSVC runtime requirement documented + Linux/Mac alternative
Stack
Python 3.10+ with PyQt6 (Qt6 runtime)
transformers + llama-cpp-python for open-weight .gguf inference on CPU
pyttsx3 for offline TTS
pynput for idle detection
Designed for low-end hardware (i7-6600U / 8GB RAM)
Why open-source AI matters here
A cloud-based version would fail this challenge: it needs internet, sends your screen-time data to a server, costs per API call, and won't run on aging hardware. Miso proves you can have a powerful, private, auditable AI companion that runs fully offline — and swap the brain with any compatible .gguf model.

Quick start

bash
git clone https://github.com/CodewithTanzeel/touch-grasschallenge-miso.git
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python scripts/download_models.py
python main.py
Windows: .\install.bat

Link:https://github.com/CodewithTanzeel/touch-grasschallenge-miso
Built with open-weight AI • Runs locally • Works offline • Costs nothing 🤖

What’s the worst productivity trap you've coded your way out of? Or, if you run Miso with a different .gguf model, let me know which one works best on your hardware in the comments below!

Tags to use in the DEV editor (Max 4):

(Make sure to only use 4 of these in the tag input fields)
Hacktoberfest OpenSource LocalAI Python

# I Built a Local AI Pet That Yells at You to Go Outside — No Cloud, No API Keys

TanzeelAhmad — Sun, 11 Oct 2026 22:07:32 +0000

Hacktoberfest Week 1 • Touch Grass Challenge

This weekend I submitted Miso — a memory-efficient desktop companion that runs a real open-weight LLM locally and literally tells you to touch grass when your focus timer hits zero.

CodewithTanzeel / touch-grasschallenge-miso

Miso — Touch Grass Desktop Companion

🤖 Built with open-source AI at its core.

What is Miso?

No cloud. Locally optimized. Memory efficient. Secure.

Miso is a cute, interactive desktop pet that sits in the corner of your screen. It works with you, not against you. The pet tracks how long you have been focused, and when your timer runs out, it gets angry (in a cute way) and literally tells you to go outside and touch grass using an open-weight language model running entirely on your laptop.

This is a Progressive Desktop Widget — fully offline, open-source, local-only, and designed specifically for low-end hardware (i7-6600U / 8GB RAM). Zero cloud APIs. Zero subscriptions. Zero data leaving your computer.

Why Open Source AI Matters Here

This project was built specifically for people with older, lower-spec hardware (like an i7-6600U with 8GB RAM). Rather than relying on expensive cloud APIs or…

View on GitHub

What it does

Miso is a cute, interactive desktop pet (Pikachu-style) that tracks how long you've been focused. When the timer runs out, it gets angry — in its cute way — and uses a local LLM (SmolLM2 / Qwen2.5) to generate a unique scolding via TTS. No cloud APIs. Zero subscriptions. Zero data leaving your machine.

Features:

Hover interaction: dance and emotions
Drag-to-position widget
Idle tracking: calms down if you actually leave
Universal customization (MISO_CHARACTER=my_pet)
Character hub: drop GIFs + config.json to build your own

The real work this week (Hacktoberfest fixes)

The PRs I submitted cover:

Deprecated param fixed
Download script repaired
Memory-efficient local AI configured
Memory file added (local/secure tracking)
.gitignore, docs, tests, TTS fallback, install errors cleaned up
User guide + emphasis on local/secure/memory-efficient
Qt6 / MSVC runtime requirement documented + Linux/Mac alternative

Stack

Python 3.10+ with PyQt6 (Qt6 runtime)
transformers + llama-cpp-python for open-weight .gguf inference on CPU
pyttsx3 for offline TTS
pynput for idle detection
Designed for low-end hardware (i7-6600U / 8GB RAM)

Why open-source AI matters here

A cloud-based version would fail this challenge: it needs internet, sends your screen-time data to a server, costs per API call, and won't run on aging hardware. Miso proves you can have a powerful, private, auditable AI companion that runs fully offline — and swap the brain with any compatible .gguf model.

Quick start


bash
git clone https://github.com/CodewithTanzeel/touch-grasschallenge-miso.git
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python scripts/download_models.py
python main.py
Windows: .\install.bat


Link:https://github.com/CodewithTanzeel/touch-grasschallenge-miso
Built with open-weight AI • Runs locally • Works offline • Costs nothing 🤖

What’s the worst productivity trap you've coded your way out of? Or, if you run Miso with a different .gguf model, let me know which one works best on your hardware in the comments below!


### Tags to use in the DEV editor (Max 4):
*(Make sure to only use 4 of these in the tag input fields)*
`Hacktoberfest` `OpenSource` `LocalAI` `Python`