SellerMate BD: Local AI Product Copy for Bangladeshi Sellers
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.
What I Built
SellerMate BD is a local AI content assistant for small online sellers in Bangladesh. Sellers enter a product name, specifications, and warranty information, then generate product descriptions, key features, SEO metadata, Facebook captions, or FAQs in Bangla, English, or both.
The problem it addresses is the time and effort required to write clear product listings, especially for independent sellers who may not have a copywriter or a budget for hosted AI services. These are the customers I built it for; I haven't yet validated demand, pricing, or willingness to pay.
Demo
There isn't a publicly deployed demo yet. The app can be run locally with Docker Compose; setup instructions are in the README. A browser recording showing synthetic data being typed, generated with local Ollama, and saved is available in sellermate-bd.
The video is a local file, not a public URL; upload it to a video host before submitting if the challenge requires an accessible demo link.
Code
GitHub repository: sellermate-bd
I couldn't identify a public repository URL for this project from the available workspace details.
One important implementation choice is the prompt's instruction to use only seller-provided facts and not invent product claims.
How I Built It
SellerMate BD uses PHP 8.4 and Apache for the web app, SQLite for local draft history, and Ollama for local model inference. The default model is qwen2.5:3b, configurable through OLLAMA_MODEL.
The app supports Bangla and English content, saves generated drafts locally, and exports them as CSV. It also includes basic tests for input validation, prompt construction, output escaping, and parsing a sample product listing.
Generated content still needs human review. This MVP has no authentication or multi-tenant isolation, so it is intended to run locally and should not be exposed directly to the public internet.
Why Does Open Innovation Matter?
Local inference gives sellers an alternative to sending product details to a hosted AI API. It avoids per-request model API charges and makes it possible to inspect and adapt the model and prompts, including experimenting with Bangla product copy.
That doesn't make AI free: users still need suitable hardware, and model licenses and terms need to be checked, especially for commercial use.
Prize Categories
I'm not entering a partner category because the project doesn't currently use a listed partner's technology.
Top comments (0)