AI
A retrieval-based assistant for order, return and product questions, with guardrails, citations and a route to a human agent.
Not client work. This sample shows how we would scope and build this type of product. Names and figures in any previews are invented.
Overview
This sample project shows how we would approach an AI support assistant for a D2C brand selling online in India; the conversations in our previews are invented. It uses retrieval-augmented generation (RAG) over the brand's help content and is designed to support the service team, not replace it.
Project facts
Illustrative sampleProduct preview
Illustrative screens. The conversation, order numbers and figures are invented.
Sample screens: names, prices and figures are examples, not client data.
How it works
The assistant answers only from the brand's own content and order data, and people stay in charge of anything involving money.
For example "Can I exchange this kurta for a larger size?", typed in English or Hinglish.
Before sharing anything about an order, the assistant confirms the phone number on the order with a one-time password.
It searches the help centre and return policy for matching passages, and fetches order and courier status if the question needs it.
The reply is written only from those passages and links to the policy it used. If nothing relevant is found, it says so.
Exchanges, complaints and refund questions go to an agent with a short summary, so the shopper does not repeat themselves.
Unanswered questions show where help articles need updating, and the evaluation set is re-run before any prompt or model change.
Then it starts again at step 1: Shopper asks a question
How we would build it
A growing D2C brand's inbox fills with the same questions: where is my order, can I exchange this size, when will my COD refund arrive. Agents copy answers from policy documents, and replies slow down during sales.
Off-the-shelf chatbots either follow rigid menus or invent confident answers. The brand needs replies grounded in its own rules and each customer's order, in English and Hinglish, with a clean hand-off to people.
Key features
7 capabilities that shape the scope and the estimate.
Platforms
A chat widget on the brand's website and a web console for the support team. WhatsApp is connected as a second channel.
Marketing and content websites: fast, search-friendly pages with a CMS your team can update without a developer.
Back-office panels for operations, support and finance teams: orders, users, content, reports and permissions.
Technology
A Next.js chat widget, a Python AI service built on OpenAI models, and a database that searches the brand's help content by meaning.
Web
A modern web technology for fast, search-friendly websites, online stores and web applications that load quickly on mobile.
Used for
Backend
A programming language for AI features, data processing and automation: the engine behind document reading, reports and smart search.
Used for
Adds AI features such as chat assistants, document summaries and smart search to your product, built on OpenAI models with human review.
Used for
Connects your product to AI language models from several providers, so each task uses a suitable model and you can switch later.
Used for
Data
A reliable database for the records your business runs on: orders, payments, bookings and stock, kept accurate and easy to report on.
Used for
Data
Keeps frequently used data in fast memory, so apps stay quick on busy days and live features like order tracking feel instant.
Used for
Cloud & DevOps
Cloud hosting for your app, website and data, with data centres in India and room to grow when traffic rises.
Used for
Privacy & security
Order details are fetched only after an OTP confirms the phone number on the order.
Tools return only the fields an answer needs; payment details never reach the model.
We confirm the provider's data-use terms and opt out of training on API data where that option exists.
Solutions
Support and sales chatbots for web, app and WhatsApp that answer from your own content and hand off to your team.
Search and question-answering over your documents, tickets and product data, with hybrid retrieval, citations and permission-aware results.
Assistants that book, schedule, look up records and raise requests by voice or text, with confirmation before every action.
Gallery
An illustrative photo of the setting this sample is built for, not a client photo.
More work
Other illustrative projects with overlapping apps, industries or services. They are samples, not client work.
E-commerce & marketplaces
An illustrative marketplace where independent fashion labels sell through one storefront, with vendor onboarding, commission rules, multi-pickup shipping, returns and scheduled payouts.
Runs on
Built with
Device management & kiosk solutions
An illustrative kiosk product for retail chains: locked Android tablets for catalogues and price checks, a staff launcher, and a console to enrol, update and monitor every device.
Runs on
Built with
Business software
An illustrative SaaS for businesses running their own vehicles: live dispatch on a map, driver proof of delivery and per-trip billing for each customer.
Runs on
Built with
FAQ
We cannot remove the risk entirely, so we reduce it in layers. The assistant answers only from retrieved passages, declines when nothing relevant is found and shows its source. An evaluation set of real, anonymised questions is re-run before any change goes live, and anything involving money, such as refunds, stays with a person.
At minimum: the help centre or policy documents, the order system (for example a Shopify store's Admin API or a custom backend) and the shipping aggregator's tracking API, plus a helpdesk such as Freshdesk or Zendesk for hand-offs. Access is read-only wherever possible, and write actions are limited to creating requests or tickets.
Mostly it depends on conversation volume. The main items are model API usage per conversation, embedding updates when content changes, WhatsApp Business Platform conversation charges and hosting. We estimate these from your current ticket volumes before the build, and can route simple questions to a smaller model. There is no fixed figure until volumes are known.
Next step
Send us the questions your team answers most, and we will say plainly which ones an assistant can handle.
