Answers from your own documents, with the source attached
An internal assistant that answers staff questions from your policies and product information, or a website assistant that handles common customer questions and hands over cleanly when it should.
- First module built in
- 3 to 5 weeks
- Answers from
- Your own documents, with sources
- Available on
- Website, WhatsApp, Teams or Slack
What this addresses
Businesses build up a great deal of written knowledge that nobody can find quickly, so experienced people spend their day answering questions the documents already answer.
- New staff spend weeks asking colleagues where things are written down
- Experienced employees are interrupted repeatedly for the same information
- Customers ask the same questions on the website and by phone
- Knowledge leaves with people when they leave
How it is put together
Your documents are indexed so an assistant can find the relevant passages and answer from them, quoting the source. It answers from your material rather than general knowledge, says clearly when it does not know, and hands over to a person when the question needs one.
What the system contains
Modules are delivered in phases, starting with the one that removes the most work. You own the finished system, code and data included.
- 01
Knowledge intake
Documents from folders, Google Drive, SharePoint, a website or a database, turned into a searchable index that refreshes as content changes.
- 02
Answers with sources
The assistant finds the relevant passages and answers from them, linking each answer to the document and section it came from.
- 03
Access control
Permissions respected, so people only get answers from documents they are allowed to see.
- 04
Handover to a person
Clear escalation to a person when the assistant is unsure or the question is outside its scope, with the conversation passed on.
- 05
Channels
On your website, inside a portal, on WhatsApp, or in Microsoft Teams or Slack for internal use.
- 06
Question log
Every question, answer and failure logged, which shows you exactly which documents need writing or updating.
Who it suits, and how it adapts
Integration effort depends on what each system offers. We assess it before quoting.
Who this suits
- Firms with detailed policies, procedures or specifications
- Schools and coaching institutes answering admission and fee questions
- Clinics answering routine patient questions about timings and preparation
- Companies whose support team answers the same questions repeatedly
- Any business where knowledge sits with a few experienced people
What it can connect to
- Google Drive, SharePoint and shared folders
- Website content and product catalogues
- Microsoft Teams and Slack for internal assistants
- WhatsApp Business and website chat for customers
- Ticketing systems for handover
How it can be customised
- Tone and response style matched to your business
- Different document sets for different user groups
- Answers in Hindi and other Indian languages
- Rules for exactly when a person takes over
- Feedback buttons on answer quality
From first call to first module
The first module is usually built in 3 to 5 weeks. Durations depend on scope, integration complexity and how quickly access and decisions are available.
- 0145 min
Free scoping call
What documentation exists, how current it is, and which questions people ask most. This decides feasibility more than any technical factor.
- 022-3 days
Scope definition
Which questions it should handle, which it must not attempt, and when it hands over.
- 033-5 weeks
Build and evaluate
Built and tested against a written set of real questions with expected answers, so quality is measured rather than assumed.
- 042-3 weeks
Internal pilot
Used internally first, even for a customer-facing assistant, so weaknesses are found before customers meet it.
AI Knowledge Assistant questions
Hosting, backups and updates after launch:
Cloud Hosting & SupportNot covered here?
Ask us directlyWill it make things up?
The design works against that. It answers from passages of your documents rather than general knowledge, cites the source so people can check, and is told to say it does not know rather than guess. It is not infallible, which is why citations and a handover path matter, and why customer-facing use should follow an internal pilot.
Our documentation is outdated. Should we fix it first?
Partly. The assistant is only as good as what it reads. But the question log is very good at showing which documents matter, so a practical approach is to start with the content that is current and let usage tell you what to fix next.
Is our data used to train AI models?
Not under the business terms of the providers we use, which exclude training on customer data. Your documents sit in your own index and passages are sent to the model only when answering a question. For highly sensitive material we can discuss self-hosted models, with an honest account of the trade-offs.
What are the running costs?
Hosting for the index and app, plus AI usage charged by the model provider for each question answered, billed to you at cost. It scales with how much people use it, so we estimate it from expected volume and set monthly limits and alerts.
Can it handle questions in Hindi?
Yes. It answers in the language the question is asked in, including major Indian languages, even when the documents are in English.
How do we know it is working?
Every conversation is logged. You can see how many questions came in, how many were answered confidently, how many were handed over and where it struggled. That log is also how it improves over time.
Around the AI knowledge assistant
Services behind it
Often combined with
Find out what your documents can answer
Share the policies or product documents people ask about most. We will test a set of real questions against them and show you the answers, with sources, before any build.
The first conversation is free, with no obligation.