Answers with citations
Every answer links to the document, version and section it used. If the documents do not contain an answer, the assistant says so instead of guessing.
Build
Every answer comes with the page it came from.
Your staff ask questions in plain language, in English or Hindi, and get answers from your policies, SOPs, circulars and manuals. Each answer cites the source document and section, and respects who is allowed to see what.
- Typical length
- 4 to 8 weeks
- Indicative price
- ₹5 lakh to ₹20 lakh, excluding GST
- Engagement path
- Discover→Prove→Build→Run
Sound familiar?
Typical situations, not quotes from clients.
- Our credit policy is 140 pages and changes every quarter. Branch staff call head office for every exception.
- New supervisors take six months to learn where anything is.
- We tried a generic AI tool. It answered confidently and quoted a policy we retired last year.
Your knowledge is spread across PDFs, SharePoint folders, email circulars and the heads of a few senior people. Search finds documents, not answers. And a tool that answers without citing its source cannot be trusted with anything that matters.
What we build
Current versions only
Retired policies are excluded. When a circular supersedes a clause, the assistant uses the new one and can show what changed.
Permissions that match yours
A branch officer sees what a branch officer is allowed to see. Access follows your existing groups in Microsoft 365, Google Workspace or your directory.
Where your people already work
Microsoft Teams, Slack, a web page on your intranet, or WhatsApp for field staff.
Hindi and Hinglish questions
Staff can ask in the language they think in, even when the documents are in English.
Feedback that improves it
Thumbs down on a wrong answer goes to a review queue, becomes a test case, and is fixed.
How it fits together
Where it fits
- SharePoint and OneDrive
- Google Drive
- Confluence
- Notion
- File servers
- Document management systems
- Microsoft Teams
- Slack
- Intranet portals
- Learning management systems
How the engagement runs
Typical length: 4 to 8 weeks from pilot start to production.
Discover
1 weekWe pick a document set and a user group, collect the 100 questions they ask most, and agree who may see what.
Prove
3 to 4 weeksA fixed-price pilot for one team on one document set. Scored against an evaluation set of real questions with answers checked by your subject experts.
Build
1 to 4 weeksMore document sets and teams, permission sync, the chosen chat channels, security review and training.
Run
monthlyWeekly review of low-rated answers, new documents indexed, evaluation re-run on each change.
How we measure it
Before we build, we agree with you in writing:
Answer accuracy
Judged by your subject experts on an evaluation set of real questions.
Citation accuracy
The cited section actually supports the answer.
Honest refusals
Questions with no answer in the documents are declined, not invented.
Permission leaks
Zero tolerance, tested on every release.
Time to answer
Compared with how long the same question takes today.
What you get
- A production knowledge assistant on the agreed channels
- Ingestion pipeline that keeps documents current
- Permission model linked to your directory
- Evaluation set and test reports, yours to keep
- Admin screen for content owners
- Runbook and training for champions in each team
- Source code and configuration in your repository
What we need from you
- The document set, with an owner for each part
- A list of the questions people actually ask (we will help collect them)
- Two subject experts for about four hours a week during the pilot
- Access to your directory groups for permissions
- A decision on which channels to support first
Indicative price
₹5 lakh to ₹20 lakh
excluding GST, from pilot to production.
The range depends on the size and messiness of the document set, the number of teams, the permission model and the channels. The pilot is a fixed price quoted after Discover. 40% advance, the rest on milestones.
Model, API and cloud costs are passed through at cost plus a 10% handling margin when we hold the accounts (no margin if you pay the providers directly). We estimate the monthly running cost before you sign.
Frequently asked questions
Is this just a chatbot on top of our documents?
The chat window is the smallest part. The work is in parsing tables and scanned pages, keeping versions current, enforcing permissions and testing answers. That is what makes it trustworthy.
Will our documents be used to train a model?
No. Your documents stay in your environment. They are retrieved at question time and are never used to train any model.
Our documents are badly scanned. Will it work?
We test a sample in Discover. Some documents need cleaning first, and we will tell you which.
Can it answer from Excel sheets and databases too?
For structured data, Analytics Agents are the better fit. Many organisations need both.
Who keeps the content up to date?
Your content owners, through a simple admin screen. The assistant picks up changes automatically.
Bring us the policy everyone calls head office about.
A 30-minute call with a founder. We will tell you what a pilot on that document set would need to prove.