This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
I was brainstorming ideas for this challenge, trying to think of something I could build that would actually help someone. Not just another project to add to my GitHub, but something that could make someone's work a little easier.
Then I overheard our cook talking about the challenges of starting a small food business.
She was worried about what would happen if the number of orders increased. How do you keep track of everything when orders come through multiple platforms? How do you make sure nobody is forgotten, deliveries aren't late, and the quality of your work doesn't suffer just because you're getting busier?
She mentioned a few customers who had already received their orders late and were understandably upset. That stuck with me. For a small business that's still finding its feet, a few unhappy customers can mean more than a couple of bad reviews. Word of mouth travels, and when your livelihood depends on people trusting you, every order matters.
I kept thinking about how much effort goes into making a small business work, and how something as ordinary as keeping track of orders can become a real problem when you don't have the right tools.
It made me wonder: what if technology could take some of that pressure off?
That's how OrderSathi started.
OrderSathi is a lightweight AI-powered order management tool for small food businesses, home bakers, cloud kitchens, and caterers who may not have the time or budget for complicated business software.
The idea is simple: instead of manually copying order details from scattered messages and screenshots, a business owner can use OrderSathi to turn them into structured orders, review the details, and keep everything organised in one place.
Here's what it can do:
- Turn messages into orders: Extract customer details, items, quantities, delivery dates, times, delivery type, and address from a natural-language message.
- Read order screenshots: Use OCR to extract text from an uploaded image and pass it into the AI extraction workflow.
- Let the user check the details: Review and edit the extracted information before saving it.
- Keep track of orders: View saved orders, update their status, and delete orders when necessary.
I wanted the tool to do more than simply generate an answer and hope it was correct. Order details affect real customers, so being able to review the result before saving it is an important part of the workflow.
OrderSathi is currently a prototype, but the problem that inspired it is real. I'd like to see whether a simple tool like this could help small food businesses handle growing order volumes without making their work more complicated.
Demo
Video demo:
The demo shows the complete journey: entering a customer's message, extracting the order using the local AI model, reviewing and correcting the details, saving the order, and updating its status. It also demonstrates how a screenshot can be converted into text through OCR.
The app isn't deployed, the video will show it running locally on my machine.
Code
OrderSathi 🍱
An AI-powered order management assistant for small food businesses.
OrderSathi helps home chefs, cloud kitchens, bakers, and small food businesses organize customer orders received through messages and screenshots. Instead of manually copying every order into a list, users can extract order details with AI, review the results, and manage orders in one place.
Built for the Hacktoberfest DEV Challenge, OrderSathi explores how open-weight AI can help simplify everyday work for small businesses.
💡 The Problem
Small food businesses often receive customer orders through informal messages and screenshots. As the number of orders grows, keeping track of items, delivery dates, timings, addresses, and order status becomes difficult.
Missing details or forgotten orders can lead to late deliveries, unhappy customers, and lost trust.
OrderSathi aims to make this process simpler by turning unstructured customer requests into organized, editable order records.
✨ Features
- AI-powered order extraction: Convert natural-language customer messages into…
The repository includes the frontend, backend, and instructions for running the project locally.
How I Built It
I built OrderSathi using:
- Frontend: React, TypeScript, and Vite
- Backend: Node.js and Express
- Open-weight AI: Qwen2.5:1.5b, run locally through Ollama
- OCR: Tesseract.js
- Storage: A local JSON file for saved orders
The main AI workflow starts with a customer's message. The backend sends it to the locally running Qwen model, which extracts the information into a structured order. The result is presented in an editable review panel, so the user can correct mistakes before saving it.
For screenshot orders, Tesseract.js first extracts the text from the image. That text can then be passed through the same extraction workflow instead of requiring the user to type everything manually.
I also implemented backend validation and order-management endpoints so that orders can be saved, retrieved, updated, and deleted. The frontend loads saved orders again after a refresh, rather than treating them as temporary UI data.
One thing I had to account for is that AI-generated details aren't always reliable. A model might misinterpret a quantity, a date, or a delivery instruction. That's why I kept the review step between extraction and saving: the AI helps organise the information, but the user remains in control.
Why Does Open Innovation Matter?
When I started thinking about this project, I kept coming back to the people it was meant to help. A tool for a small business shouldn't introduce another expensive recurring bill just to perform a routine task.
That's one reason I chose to use an open-weight model locally through Ollama.
Once the model and required dependencies are installed, OrderSathi can perform its AI extraction without making a paid, hosted AI API request for every order. That removes per-request charges from the AI provider and gives the project a path towards keeping routine AI processing affordable.
There is also a privacy consideration. Order messages can contain names, phone numbers, and delivery addresses. Running inference locally can reduce the need to send that information to an external AI service. It doesn't automatically make the entire application private or secure, but it gives me more control over where the AI processing happens.
Open-weight models also give developers room to experiment. I can adjust the extraction instructions, test another compatible model, and explore whether a different model works better for food-order language without having to redesign the whole application around one provider.
For me, that's what open innovation makes possible: the ability to experiment, learn, and build something useful without needing access to a large budget or a proprietary AI platform.
I'd rather start with a small tool that solves one problem and improve it through real feedback than build something complicated that its intended users never need.
What I'd Like to Improve Next
OrderSathi is a working prototype, but there's still a lot I want to improve. My next priority is voice-note parsing, so food business owners can turn audio orders into structured details without having to listen repeatedly and type everything manually.
I'd also like to make the AI more reliable when customers provide incomplete or ambiguous information, improve how missing details are highlighted, and make order management easier as the number of orders grows. In the longer term, I want to explore a more robust database instead of local JSON storage and test the app with more small food businesses to understand what they actually need.
For now, my goal is to keep the tool simple, affordable, and genuinely useful—building improvements around real feedback rather than adding features just for the sake of it.
My First User
My first user was our cook, the person who inspired OrderSathi. She tried the prototype and gave it a positive review, especially appreciating that she could upload screenshots of customer orders instead of having to copy and paste every message manually. She also said she would happily pay a small one-time fee for a tool that helped her organise everything in one place, easily. Her suggestion for the next improvement was voice-note parsing, since customers often send orders through audio messages.
That feedback made the project feel a little more real to me. My next step is to add voice-note processing, turning spoken orders into structured, editable details just like text messages. It's encouraging to know that something I built to address a problem she described could be useful enough that she'd consider paying for it.
Prize Categories
Best Use of ElevenLabs: I used ElevenLabs to generate AI voiceover narration for the OrderSathi demo video, explaining the application's features and workflow. This made the demonstration clearer and more engaging while showcasing how AI-generated speech can help communicate a project's purpose and functionality.
A Small Beginning
OrderSathi began with a conversation I wasn't even part of. It reminded me that useful projects don't always start with a big idea. Sometimes they start with noticing a problem that someone around you is struggling with and asking whether you can do anything to help.
I don't know yet how far this project will go. But I hope it can become something practical for the people it was built for, and that this is a first step towards building technology that makes everyday work a little easier.
Top comments (0)