This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
🛍️ DukaanDraft — Your products. Better words. More sales.
Small business owners often spend more time writing product descriptions than actually selling their products. Creating separate captions for Instagram, descriptions for WhatsApp catalogues, promotional messages, and answers to customer questions can quickly become repetitive.
I built DukaanDraft, an AI-powered listing assistant that helps small Indian sellers turn basic product details into a ready-to-use marketing pack.
The idea is simple: enter your product name, price, features, preferred platform, language, and tone. DukaanDraft generates:
- 📸 Instagram captions with relevant hashtags
- 💬 WhatsApp catalogue descriptions
- ✨ Three alternative opening hooks
- ❓ Frequently asked questions based on the supplied product facts
- 📣 A short promotional message
It supports English, Hindi, and Hinglish, with different tones for different kinds of businesses.
The goal is to help independent sellers, home businesses, jewellery makers, and small online stores spend less time writing copy and more time running their businesses.
Demo
Try DukaanDraft:
For the demo, I used an organic Kachi Ghani yellow mustard oil listing priced at ₹349 and asked the model to generate WhatsApp marketing copy in Hinglish.
The generated content preserved the supplied product details, including the Rajasthan yellow mustard seeds, traditional wooden kohlu extraction below 45°C, and BPA-free UV-protected amber bottle.
It also handled missing information correctly. When asked about cash on delivery and shipping timelines, it indicated that the seller needed to confirm those details rather than inventing an answer.
Code
GitHub repository: https://github.com/Tanishaaaaaaa/dukaan-draft
The project includes the frontend, backend generation endpoint, model integration, and product-copy generation workflow.
How I Built It
I wanted to keep DukaanDraft small enough to build quickly but useful enough to solve a real business problem.
The application uses:
- React and Vite for the user interface.
- Node.js and Express for the backend API.
- Google's open-weight Gemma model for product-copy generation.
- Google's Gen AI SDK to send prompts to the model.
- Environment-based secret management to keep API credentials out of frontend code.
The frontend collects product details and sends them to a backend /api/generate endpoint. The backend constructs a prompt using the seller's inputs, calls the configured Gemma model, and returns the generated content to the interface.
The application is configured to use gemma-4-26b-a4b-it. The generation flow was tested end to end, with a reported successful response and approximately 45 seconds of inference latency during the test.
I also added error handling with retries for transient upstream failures, individual copy actions for generated FAQs, and responsive layouts for mobile screens.
One important design decision was to ground the output in the seller's actual product information. Marketing copy is useful only when it remains trustworthy. DukaanDraft should not invent prices, product attributes, shipping promises, or answers to questions the seller has not addressed.
Why Does Open Innovation Matter?
For a small side hustle, the economics of AI matter.
A seller generating a handful of product listings should not need an expensive, complicated software stack just to write product descriptions.
Using an open-weight model such as Gemma gives developers the flexibility to experiment with different model configurations, tailor prompts to a specific audience, and explore different inference setups as a product grows.
For DukaanDraft, this creates several opportunities:
1. Building for a specific audience
Generic writing assistants are designed for a broad range of tasks. DukaanDraft focuses on the needs of small sellers, including platform-specific copy, INR pricing, and conversational Hinglish marketing.
2. More control over the AI workflow
The generation prompts and output structure can be adapted to the seller's business without rebuilding the application around a proprietary, closed model.
3. Flexibility in deployment and cost
Open-weight models provide options for hosted inference or, where supported hardware and model requirements permit, self-hosted inference. Comparing these options could help reduce costs as usage grows and give the application more control over its infrastructure.
4. Building without unnecessary complexity
I could focus on a narrow, useful workflow instead of building a large AI platform. The model performs the language-generation task, while the application handles product inputs, formatting, and copying the results.
There are trade-offs, too. Hosted inference still depends on provider availability, latency, and applicable usage limits. Running an open-weight model does not automatically make inference free or guarantee that customer data remains private.
For this version, I used hosted Gemma inference. A future version could explore local inference for sellers who need greater control over their data.
To me, open innovation means having more choices about how an AI product is built, customized, and operated.
My Agent Session
The project was developed and tested with an AI-assisted coding workflow in Google AI Studio.
I used iterative prompts to build the application, integrate Gemma, and audit the connection between the frontend and backend. The audit also covered product-fact preservation, API error handling, copy actions, and mobile responsiveness.
I did not save a shareable DevRelay session for this build, so I am not including a session link.
Prize Categories
- Gemma — Best Use of Gemma: DukaanDraft uses Gemma as its configured product-copy generation model.
- Overall Hacktoberfest Open-Source AI Challenge: A focused AI product designed around a real customer problem and a plausible path to monetization.
What's Next?
The next step is to put DukaanDraft in front of real small-business owners and learn whether it genuinely saves them time.
I'd especially like to test it with independent sellers who manage their own Instagram and WhatsApp catalogues. Their feedback will help determine which features are worth building next and whether a paid tier for generating listings in batches makes sense.
Because a useful side hustle should start with a real problem, not just an interesting model.
Thanks for checking out DukaanDraft! 💗
Top comments (2)
Hey i watched your video on YT, Your Project is nice i want a bit of a help can you give mesome advice about which AI to use and how to submit my project
Hi, yes sure I can help you with the same. Let me know your project details for further clarity on the same.