DEV Community: V Himacharan

MENUMATE-AI

V Himacharan — Mon, 19 Oct 2026 16:11:59 +0000

MenuMate AI: A Local-AI Restaurant Menu Translator 🍽️🌿

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

What I Built

I'm a first-year Computer Science Engineering student, and I built MenuMate AI, a multilingual restaurant menu translator powered by a locally running AI model.

Restaurant owners and small cafés may need to make their menus understandable to customers who speak different languages. Professional translation can add time and cost, especially for small businesses.

MenuMate AI lets users enter a restaurant name, dish name, and description, select a target language, and generate a translation using local AI. Users can copy the translated text or download it as a text file.

Features

Multilingual menu translation
Local AI inference using Ollama and Qwen2.5:1.5B
A clean, restaurant-friendly interface
Copy and download options for translated menus
Customizable restaurant name

My goal is to explore a practical tool that small restaurants could use to prepare multilingual menu content. This is an early prototype, and translations should be reviewed before being used commercially.

Demo

Demo video:https://drive.google.com/file/d/1ZxoJWzaPJX9U50WThRvGWyBx586lIfkp/view?usp=drivesdk

The application currently runs locally on my Windows laptop. I haven't deployed a public live version yet, so I'm sharing a demo video rather than claiming the local URL is publicly accessible.

Code

GitHub repository: https://github.com/vyshu1230/Menumate-AI

The repository contains the project code, dependencies, and setup instructions.

How I Built It

I built MenuMate AI using:

Python for the application logic
Streamlit for the web interface
Ollama to run the model locally
Qwen2.5:1.5B as the language model

The workflow is straightforward:

The user enters a restaurant name and menu details.
The user selects a target language.
The application sends the translation request to the local model through Ollama.
The model generates the translated description.
The interface displays the result and provides copy and download options.

I tested the application locally and verified that the translation interface works.

Why Does Open Innovation Matter?

Open innovation made this project possible as a beginner-friendly experiment.

Using an open-weight model through local inference gave me a way to build an AI-powered application without depending on a hosted AI API for every translation request.

It also gave me the opportunity to learn how a language model connects to a real application, how to design a useful interface around AI, and how to test the complete workflow.

Local inference can offer more control over deployment and help reduce dependence on external inference services. However, it still requires suitable hardware, and model quality and speed depend on the computer and model being used.

I believe open-source tools and open-weight models give students and small developers more opportunities to experiment, learn, and build practical solutions.

Prize Categories

I'm entering for the overall Hacktoberfest Open-Source AI Challenge. I'm not claiming an additional partner prize category because this project currently uses Ollama and Qwen2.5:1.5B rather than a verified challenge partner technology.

Team

@vyshu1230 :Error debugging
@pashya_rishitha_0610 :Designed and build the app|Language translation|My restaurant.
@himacharan69:Helped to design the App

AI assistance disclosure: I used AI assistance while preparing this article. I reviewed the content against my project and testing experience before publication.

hf26challenge #python #ai #showdev #beginners #ABotWroteThis

Trail Sense("From Screen to soil")

V Himacharan — Mon, 12 Oct 2026 14:42:19 +0000

**This is a submission for the [Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass]

What We Built

TrailSense ("From Screen to Soil") is a beginner garden helper for people who want to grow something but don't know where to start. It is for students and first-time gardeners with a windowsill or balcony.

It is built so the screen is the shortest part of the experience:

Find Plants: you set your space, sunlight, budget (in rupees) and daily care time. The app scores each plant against your setup and explains why it fits.
My Garden: save plants, get a small daily task for each one, mark tasks complete, and remove plants. Your garden is saved in a local file.
AI Assistant: ask a gardening question and get a short answer in simple words. Every answer ends with one 5-minute outdoor task to do away from the screen.
Today's little challenge: the dashboard nudges you to step outside for five minutes, notice the sunlight, and feel the soil.

Team

@vyshu1230 : Designed and built the app: plant matching, My Garden, dashboard.
@himacharan69: Found that the AI Assistant page was missing and built it, tested the app.
@pashya_rishitha_0610 : Helped to design the app.

Demo

The app runs locally, so there is no hosted link. That is the point: it needs no server and no signal.

1)
!(https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/pva141e5zphxwrxb5b7i.png)

2)
!(https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/pvfed0cgi4zlttqq6eg7.png)

Code

https://github.com/vyshu1230/Trailsense-/invitations

How We Built It

Model: llama3.2, an open-weight model, run locally with Ollama.
App: Streamlit and plain Python. Plant data lives in plants.json, and your saved garden in garden_progress.json.
Plant matching: a scoring function rates sunlight fit, space fit, daily care time and cost. Budget is a hard limit.
AI Assistant: the model gets a short instruction, plus your sidebar setup and the plants you have saved:

You are TrailSense, a friendly gardening helper for beginners.
Use the garden setup you are given. Answer in simple words, in at most
5 sentences. Do not invent facts about plants; if you are not sure, say
so. End with one short outdoor task the person can do in 5 minutes away
from the screen.

The first version of the app listed an AI Assistant in its menu, but the page was never built. We found this while testing and added it.

Where it falls short:

Only five plants (Tulsi, Mint, Aloe Vera, Tomato and Spinach) have their own daily task. Every other plant gets the same generic task.
A small local model can make mistakes, so its advice is a starting point, not expert advice.
A vague answer or an ignored setting.

Why Does Open Innovation Matter?

A garden helper is most useful where there is no signal: a balcony, an allotment, a trail. A closed API would need internet and an account in exactly those places.

It works without a connection. The model runs on the laptop. We turned off Wi-Fi and the assistant still answered.

!(https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/9c0f1pt91i652fy66s8y.png)

Your data stays with you. What you plant and ask never leaves your machine.
It costs nothing to run. There is no API key and no bill.
We can swap models. Changing one line (AI_MODEL) switches to any model shown by ollama list.

One honest note: llama3.2 is open-weight, not fully open source. We can download and run it, but its training data is not public.

What's Next

We have not yet tested TrailSense in a real garden. Our plan is to take it outside, ask the assistant about the plants in front of us, do its 5-minute task, and report how it goes.

My private Day planner

V Himacharan — Mon, 05 Oct 2026 18:52:54 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What We Built

Shruthi is a close friend of all three of us. She told us that [HER REAL PROBLEM, IN HER OWN WORDS, e.g. "my days feel messy and I never stop to reflect"]. She also said she would never type her personal thoughts into an online app.

That last sentence shaped everything. So we built a private day planner that runs fully on a laptop, with no account, no API key and no server.

It has two tabs:

Morning plan: she types her day ("college 9 to 4, assignment due tomorrow, I want to exercise and sleep by 11") and gets an hour-by-hour schedule.
Night reflection: she writes a few lines about her day. The app saves them to a text file on the laptop, then gives a short summary, one thing that went well, and a small goal for tomorrow.

Team

@himacharan69: wrote the first version and the system prompt.
@vyshu1230 : built the Night reflection tab.
@pashya_rishitha_0610 : model testing and screenshots.

Demo

The app runs locally, so there is no hosted link. That is the point: her diary never leaves the machine.

: Morning plan tab with a schedule.

: Night reflection tab with the saved message

What shruthi Said

I want a proper private dairy which I used check that often Ly and it is very usefull to me the main thing i want is it should be in my device.

Code

https://github.com/Himacharan69/private-day-planner.git

How We Built It

Model: llama3.2, an open-weight model, run locally with Ollama.
Interface: Streamlit, so the whole page is plain Python.
Storage: diary entries are appended to a diary.txt file on the laptop.
Hardware: a normal Windows laptop with 32 GB of RAM.
Learning along the way: we used an AI assistant to learn Ollama and Streamlit and to debug errors.

The most important part turned out to be one short instruction (a system prompt) that tells the model how to behave:

You are a kind day planner. Use ONLY the times and tasks the user gives.
Do not change fixed times like classes. Fill the free time with study,
meals, exercise and rest. Format as a markdown list with each time slot
on its own line. End with one short encouraging sentence.

To see how much it mattered, we ran the same input ("College 9 to 4, assignment due tomorrow, I want to exercise and sleep by 11") two ways:

Plain terminal chat, no instructions: llama3.2 ignored the college hours, wrote "11:00 PM" twice, and invented a 6 AM wake-up.
Inside our app, with the system prompt: it kept college from 9 to 4, fitted in the assignment, exercise and dinner, and ended with sleep at 11.

Same model, very different result.

Where it falls short: a small model still makes mistakes. [ONE REAL PROBLEM YOU SAW, e.g. an invented task or a strange time slot.]

Why Does Open Innovation Matter?

A diary is the most personal thing a person writes. On a closed API, every entry would travel to someone else's server, and the app would need internet and a paid account. With an open-weight model:

Her data stays with her. Nothing is sent anywhere.
We could change the model's behavior ourselves. One short prompt fixed the biggest problems in minutes, for free.
We could swap models. Trying a different one meant changing a single line of code.
It costs nothing to run.

One honest note: llama3.2 is open-weight, not fully open source. We can download and run it, but the training data is not public.

What's Next

We want the morning plan to read yesterday's diary goal, so the two tabs become one daily loop. Building this for someone we all know taught us more than any demo project would have.