MAKIN AI chatbot development that answers from your own documents
What’s included
01- ✓Document ingestion and chunking strategy
- ✓Vector store with permission filtering
- ✓Citations on every answer
- ✓Evaluation set before go-live
- ✓Refusal behaviour for unknowns
- ✓Usage and cost dashboards
The details
02How is this different from a general AI chatbot?
A general chatbot answers from whatever its model learned in training, which is why it can sound confident and still be wrong about your business. This assistant uses retrieval-augmented generation (RAG): each question first searches your own documents, and the answer is written from the passages it finds, with citations so the reader can open the source and check.
When nothing relevant turns up, its refusal behaviour is designed to say it does not know rather than fill the gap. For policies, product specifications, SOPs and contract terms, that is the point: a wrong answer delivered confidently costs more than an honest ‘not in the documents’.
Who can see which documents?
Permission filtering is built into the vector store, so a search only looks at documents the person asking is allowed to see. A leave policy can be open to everyone while salary bands stay with HR; a client file can stay with the team working on it. Which access rules apply, and where they come from, is agreed on the scope call and written into the proposal before any document is ingested.
How is it tested before people rely on it?
Before go-live we build an evaluation set: real questions your people ask, the answers they should get and the documents those answers live in. The assistant is tested against it, and the chunking strategy — how documents are split so the right passage is found — is adjusted until the results hold up. After launch, the usage and cost dashboards show how much the assistant is used and what it costs to run.
What is not included, and when is another package better?
Model usage is paid to the provider on top of this price. Preparing your source material — scanning paper files, retiring out-of-date policies — is not part of the package; the assistant can only be as current as the documents it is given. The interface it appears in is scoped on the call.
If you want customers on WhatsApp answered and routed with a human hand-off, WhatsApp & Support Automation starts at ₹12,000. Support teams who want reply drafts built from their resolved tickets have the AI Support Copilot, from ₹45,000. If the questions are about your Tally books, the Tally AI Assistant starts at ₹22,000.
How it works
03- 0115-min callTell us what’s breaking. We ask the questions that change the price.
- 02Scope in writingA fixed quote and a written scope: what’s in, what isn’t.
- 03Build & reviewWorking software early, with review rounds on real screens.
- 04LaunchWe go live with you, train your team and hand over access.
- 05SupportFixes, changes and growth on a retainer, if you want one.
FAQ
04What does AI chatbot development cost?
The Enterprise RAG Knowledge Assistant starts at ₹1,75,000 and is delivered in up to 3 months. That covers document ingestion, the permission-filtered vector store, citations, refusal behaviour, the evaluation set and usage and cost dashboards. The model provider’s usage charges come on top. Your fixed price arrives in writing after a 15-minute scope call.
Is the chatbot trained on our data?
Not in the sense of retraining a model. Your documents are ingested, split into passages and stored in a vector store; when someone asks a question, the relevant passages are retrieved and the model answers from them. Updating an answer usually means updating the document, not retraining anything.
What happens when it doesn’t know the answer?
It is built to say so. Refusal behaviour for unknowns is part of the package: if your documents don’t contain the answer, the assistant declines rather than inventing one. The evaluation set can include questions whose answers are deliberately missing, so this behaviour is tested before launch rather than discovered by a user.
Which documents can it use?
Text-based material your team relies on: policies, manuals, SOPs, product sheets, contracts, past proposals. The ingestion and chunking strategy is designed around your document types, because a long contract and a short FAQ need splitting differently. The document types, formats and volumes are settled on the scope call.
Can it answer customers on our website or WhatsApp?
Where the assistant appears is scoped on the call rather than fixed in the package. If the job is mainly customer questions on WhatsApp — order status, routing, hand-off to a person — WhatsApp & Support Automation from ₹12,000 is the better fit. This assistant is for questions that need your documents and a citation.
More in AI
0515 minutes.
One quote.
Tell us what you need. We reply on WhatsApp or email with a time, and a written quote after the call.
Re: Enterprise RAG Knowledge Assistant