The Party Shop | Riyan Mart: Bangla AI Sales Assistant for WhatsApp and Facebook Messenger, Connected to Live WooCommerce Orders

Built and delivered a custom AI sales assistant that answers customers in Bangla on WhatsApp and Facebook Messenger, searches the live WooCommerce catalogue, takes real orders, reports order status and hands over to the support team when a person is needed. Delivered between 14 and 25 October 2026.

Client: The Party Shop | Riyan Mart. Industry: E-commerce & Party Supplies. Location: Dhaka, Bangladesh.

The challenge

The Party Shop | Riyan Mart (partyshopbd.com) sells party supplies online and receives a constant flow of customer questions on WhatsApp and Facebook Messenger: which product fits an occasion, price and stock, delivery charge and time, and where an order is. The shop needed an assistant that follows its own written sales rules, answers in Bangla and never invents prices or promises.

  • Replies slowed down whenever many chats arrived at once or the team was offline.
  • Sales rules (discount limits, delivery charges, same-day delivery conditions, wording for special requests) had to be remembered and applied consistently by every team member.
  • Complaints, wholesale, custom print and refund requests needed to reach a person quickly without the AI improvising an answer.

The solution

NextCore BD customised its own AI sales platform for this shop: a rule-driven assistant connected directly to the shop's WooCommerce store, with a CRM inbox, a monitoring page and a complaint box so the owner can run and correct it without a developer.

  • Translated the shop's written requirements (36 sales and safety rules plus a revised sheet of 20 changes) into prompt rules and hard-coded checks, for example fixed wording for common requests, discount ceilings the customer is never told about, and one follow-up message after five minutes of silence.
  • Order flow: summary first, order created only after a clear confirmation, delivery charge taken from the shop's WooCommerce shipping settings, expected delivery date calculated from the shop's delivery rules.
  • Order status lookup by phone number or order number, with hand-over to the support team whenever the order is not found or the date has passed.
  • Complaint, refund, wholesale, custom print and combo customisation requests are handed to the team with the exact agreed message; the AI never promises refunds or prices for custom orders.
  • Steadfast courier booking was built and kept switched off on purpose until the client supplies its API keys.
  • Independent review of the full code, then live AI scenario testing and browser testing before hand-over, plus a written user guide and progress reports for the owner.

Verified outcomes

  • The assistant replies in Bangla on WhatsApp and Facebook Messenger and searches the live WooCommerce catalogue of about 290 products, including photo matching.
  • Orders taken in chat are created as real WooCommerce orders, with the AI channel shown in the Origin column.
  • Customers can ask for order status and expected delivery date; unclear cases go to the support team instead of the AI guessing.
  • The owner runs the system from a CRM inbox, a monitoring page and a complaint box, with e-mail and push alerts.
  • At hand-over: 191 automated tests passing, 29 main pages checked in a real browser, and 73 live AI scenarios run three times each (about 99% of checks passed, with failures fixed afterwards).

Technologies

Python / Django, WooCommerce REST API, WhatsApp (Evolution API), Facebook Messenger (Meta Graph API), OpenAI / OpenRouter, ChromaDB Vector Search, OpenCV & Tesseract OCR, PostgreSQL, Docker & Dokploy