I built ZVYA AI: a local-first business assistant for small sellers
What I built
I built ZVYA AI, a local-first AI business assistant for small sellers, independent makers and home businesses. Many small businesses receive orders as informal chat messages, then manually copy product details, quantities, prices and delivery requests into notes or spreadsheets. ZVYA AI helps turn those messages into structured orders and keeps the seller in control of the final record.
A seller can paste a customer message, ask a local AI model to extract order details, review and edit the result, then save the confirmed order. The app also supports order status updates, manually recorded payments, outstanding balances, business summaries and CSV export.
Code
GitHub: [https://github.com/ashandeepkaur27/ZVYA---AI.]
How I built it
- Python + Streamlit for the app and interface.
- Ollama + Gemma 3 for local extraction of order details from customer messages.
- SQLite for storing orders locally.
- Pandas for lightweight reporting and CSV export.
The app does not need a hosted AI API key. In the default setup, the prompt is sent to the Ollama service on the same computer. The extracted fields are shown for review before the seller saves an order. This human-review step matters because AI can misunderstand prices, dates or quantities.
I also kept a distinction between an advance amount mentioned in a customer message and a payment that the seller has actually received. Only manually recorded receipts count toward the payment summary.
Why this matters
Small sellers deserve practical digital tools without needing a complicated subscription-based business platform. ZVYA AI explores how a local AI model can reduce repetitive order entry while keeping data and the final decision with the seller.
What I learned
Building this project helped me connect a local language model to a working Streamlit interface, validate model output, persist records with SQLite, and think about the difference between AI-extracted information and verified business data.
What's next
Potential next improvements include customer follow-up message drafts, reminders for upcoming delivery dates, and more robust date parsing.
Top comments (0)