Book discovery call
Home / Services / AI Chatbot Development
AI chatbots

AI Chatbot Development that answers from your own documents

A chatbot for your website or WhatsApp that answers from your own PDFs, price lists and policies, and hands over to a person when it is unsure.

A useful AI chatbot is not a general chat window with your logo on it. It is a system that finds the right passage in your documents and answers only from that. When it does not know, it says so and passes the customer to your team. That design choice is what separates a helpful bot from an embarrassing one.

We build these chatbots using retrieval over your own content: brochures, fee structures, product catalogues, FAQs, policy documents and past support answers. They run on your website, on WhatsApp, or both, and respond in English, Hindi or Marathi depending on the customer. You decide which topics it may discuss and which it must pass to a person, and we write those limits down before building.

Why most business chatbots disappoint

01

Answers that sound right but are wrong

A model left to improvise will invent a refund policy or a delivery date. For a business this is a liability, so every answer must be tied to a source you control.

02

Scripted bots that frustrate users

Old button-menu bots break when someone types a normal question. Customers abandon them and call you anyway, which defeats the purpose.

03

No plan for the handover

When a bot fails, the customer needs a person, with the chat history attached. Without this, the enquiry is lost or the customer repeats everything.

04

Unclear running costs

Model usage is billed per use. A bot with no limits, caching or monitoring can produce a surprising monthly bill once traffic grows.

What we build when we build an AI chatbot

01

Document-grounded answers

We index your PDFs, web pages and spreadsheets, split them sensibly and retrieve the relevant parts for each question. Answers show their source so your team can check them.

02

Website and WhatsApp deployment

A chat widget on your site, a WhatsApp Business number, or both from the same knowledge base. WhatsApp setup includes template approval and opt-in handling.

03

Guardrails and boundaries

The bot is limited to topics you approve, refuses requests outside scope, avoids giving legal or medical advice, and never reveals internal instructions. We test it with deliberately hostile questions.

04

Human handover

When confidence is low or the customer asks for a person, the chat moves to your team with full history, by email, WhatsApp or a simple inbox. Out-of-hours messages are queued.

05

Lead and enquiry capture

The bot can ask for name, phone and requirement at the right moment, then push that to your CRM, Google Sheet or email. It avoids asking for details before it has helped.

06

Admin panel and review

You upload new documents, see which questions were asked, mark bad answers and correct them. Weekly review of unanswered questions shows what content is missing from your site.

When this is the wrong choice. If customers mainly ask five fixed questions, a short FAQ page or a simple WhatsApp auto-reply is cheaper and safer. An AI chatbot also needs good source material. If your policies are not written down or keep changing, fix that first.

How we take an AI chatbot from idea to live

01

Discover

We read your documents and past customer chats to find the real top questions. We also decide which questions the bot must refuse or hand over.

02

Define

We agree on the source documents, tone, languages, handover rules and a test set of fifty to one hundred real questions with correct answers.

03

Design

Conversation flow, widget look and WhatsApp message format are designed together so the bot feels the same on both channels.

04

Build

We build retrieval, answer generation and guardrails, then score the bot against the test set. Accuracy is measured before launch, not guessed.

05

Ship & extend

We launch to a small audience first, review live conversations weekly and tune. Support runs for one month, and new channels or documents can be added later.

OpenAI APIAnthropic Claude APIGoogle Gemini APIPythonNode.jspgvectorPostgreSQLLangChain or custom retrievalWhatsApp Business Cloud APIReactRedisDocker

What it costs

Starts at ₹1,00,000 (excl. GST)

That covers a document-grounded chatbot on one channel with handover and an admin panel. Extra channels, many languages, CRM integration and large document sets are quoted after a discovery call.

  • Volume and format of source documents
  • Website only, or WhatsApp as well
  • Languages supported
  • CRM or helpdesk integration
  • Monthly model usage, billed at actual provider cost

All prices exclude 18% GST. See also website costs and app costs.

Questions we get asked

Will the chatbot make up answers?

We design it to answer only from retrieved passages and to say it does not know otherwise. No system is perfect, so we test against real questions, show sources, and give you a review screen to correct errors.

Can it work on WhatsApp?

Yes, through the WhatsApp Business Cloud API. You need a business number and Meta verification. Replies inside the 24-hour window are free-form, and messages outside it need pre-approved templates, which we help set up. Accuracy is lower for rare words and mixed scripts, so testing on your real chats matters.

Does it understand Hindi and Marathi?

Current models handle Hindi and Marathi reasonably, including Hinglish typed in Roman script. We test with your customers' actual phrasing, and tell you where accuracy falls short before launch. You can also choose a region and a provider that suits your internal data rules.

Where does our data go?

Document text and questions are sent to the model provider you choose to produce answers. Providers offer settings that stop your data from being used for training. We document this flow so you can decide if it meets your needs.

What does it cost to run each month?

Model usage is billed by the provider based on how many questions are asked, plus hosting. For a small business it is usually modest, and we add limits and caching to keep it predictable. We estimate it in the quote.

Can it book appointments or take orders?

Yes, if it is connected to your calendar or order system. That makes it an agent that takes actions, so we add confirmation steps and limits. We usually launch with answers first and add actions after trust is built.

Ready to scope it?

Tell us what you want to build. You get a fixed written quote after a free discovery call.