BazaarMate AI

An AI sales copilot for Afghan small businesses that turns Dari and Pashto customer messages into professional replies, structured orders, follow-ups, and sales insights.

Comment

Inspiration

Many small businesses in Afghanistan and other emerging markets sell through WhatsApp, Facebook, and Instagram instead of traditional online stores. Customers send informal messages in Dari or Pashto containing product questions, quantities, phone numbers, delivery locations, and cash-on-delivery requests. Merchants must manually interpret every message, write a reply, and transfer the information into an order list. Important details are easily missed and slow responses can mean lost sales.

We built BazaarMate AI to turn these conversations into clear, actionable orders while respecting local language and buying habits.

What it does

BazaarMate AI is a mobile-first sales copilot for social-commerce merchants. A seller pastes a customer message into the inbox and the application:

  • Detects the language and purchase intent
  • Extracts the customer, product, quantity, phone number, and delivery location
  • Identifies missing information and recommends the next action
  • Drafts a culturally appropriate reply in the customer’s language
  • Converts the conversation into a structured order
  • Tracks the order from New to Confirmed to Shipped

The interface is designed for merchants who work primarily from a phone and depend on cash-on-delivery workflows.

How we built it

We used Codex and GPT-5.6 to design, implement, review, and test the complete prototype. The application is built with TypeScript, React, Next.js, Vinext, and Cloudflare Workers, and is deployed through OpenAI Sites.

The server-side analysis route uses the OpenAI Responses API with GPT-5.6 Terra. It asks the model to return structured JSON containing the detected language, confidence, customer information, order fields, intent, localized reply, and follow-up action.

The API key remains server-side and is never exposed to the browser. The public prototype also includes a deterministic Dari demonstration fallback, allowing judges to test the complete message-to-order workflow if a live model request is unavailable.

Orders created in the demo are stored locally in the browser. The project includes responsive mobile design, a public GitHub repository, documentation, automated linting, build validation, and a working deployed demo.

Challenges we ran into

One major challenge was preserving the meaning and tone of informal Dari messages while extracting structured business data. Customer messages often combine names, product descriptions, addresses, quantities, and questions in a single sentence.

We also needed to distinguish confirmed information from missing information so the system would not invent personal data. Another challenge was creating an interface that remained simple and readable on small mobile screens while still showing order details and follow-up actions.

Finally, we designed a safe fallback workflow so the prototype remains demonstrable without exposing credentials or failing completely when an external model request is unavailable.

Accomplishments that we are proud of

We created a complete working flow from an unstructured Dari customer message to a structured, ready-to-confirm order. BazaarMate can generate a localized response, capture delivery information, create an order, and advance its status through the seller workflow.

We are also proud that the project addresses a real gap: most commerce tools are English-first and assume customers use structured checkout forms. BazaarMate starts with the way people already communicate.

What we learned

We learned that language is not simply a translation layer; it is part of the product experience. A useful response must preserve local tone, commerce practices, and context.

We also learned the importance of structured model outputs, server-side credential handling, deterministic fallbacks, human confirmation, mobile-first design, and clear documentation for building reliable AI workflows.

What’s next for BazaarMate AI

Next, we plan to add merchant accounts, encrypted persistent order storage, stronger Dari and Pashto evaluations, catalog-aware product matching, voice-message transcription, and integrations with WhatsApp and Facebook.

We also want to add human approval controls, delivery coordination, follow-up reminders, and practical sales insights while keeping the product simple enough for small merchants to use from day one.

Built With

Share this project:

Updates