Policy questions for a growing team
Staff ask about leave, expenses or travel instead of searching a drive. The answer shows the policy passage and says so when the match is weak.
In short. Aiwah Labs in Bengaluru builds company brains: internal AI knowledge bases that answer staff questions from your own documents and show the source. The first use case is live within a week, and we build on it. Typical setup is from about ₹1,25,000, with no monthly maintenance fee. You own the code and the data.
Company brain implementation overviewAll services in Bengaluru

A company brain is an internal system that answers staff questions from your own documents and records. It shows the source, and it refuses what the asker may not see. It prepares answers. A person still approves anything that changes a record.
Uploading a folder to a public chatbot makes a private copy outside your control, under that product’s retention rules. A company brain runs inside your operation, with permissions, a source for every answer, and a store you own. For the wider controlled workflow, see knowledge and back office.
Only the sources named in the engagement are connected.
Access follows the source system where it exposes a list.
The passage used is shown. A weak match is reported as weak.
Nothing that changes a record runs without a person.
Scenarios the setup is built around. Each one ends in a record or a person, not a dead end.
Staff ask about leave, expenses or travel instead of searching a drive. The answer shows the policy passage and says so when the match is weak.
An on-call engineer asks how a past outage was handled. The brain points to the runbook and the incident note, within what that person may open.
Finance asks what was agreed with a supplier. The brain finds the clause, shows the page, and leaves the decision to a person.
We connect the sources your team uses today, as long as each one has an interface. We agree the list before we build.
Google Drive, SharePoint, OneDrive, Confluence, Notion, email exports, or the internal database your team treats as the record. The engagement names the list, and it stays the list.
Each answer cites the passage it used. When the source is missing or the match is weak, the system says so.
Where the source system exposes an access list, the answer follows it. Where it does not, we scope a narrower source instead of guessing.
The index sits in your own cloud account, in a region you choose. Staff ask in Slack, Teams or a simple web page. The models that draft the answer can be swapped later. The files stay yours.
How much notice do I need to give for unpaid leave?
Two weeks. Section 4.2 of the Leave Policy says to submit the HR form and your manager approves it.
Leave Policy, section 4.2. Answered within this person’s access to the HR folder.
And what is the notice for the sales commission plan change?
I cannot find that in the documents you may open. I have asked the finance lead to answer.
Answers follow the folder access each person already has.
First use case in 1 week. We build on top of it, one use case at a time.
We list the repeated questions, the folders they should come from, and who may see each folder.
We trace who can open each source today, so the brain never becomes a side door around it.
Only the named sources get connected. Every answer keeps a link back to its passage.
Real questions get scored for the right source, a correct refusal and the right permission. That sample is the accuracy check.
Staff ask inside the tool they already open. The first use case is live within a week.
The next use case goes on top of the first, one at a time, with the same source for every answer and the same permissions.
Prices are in Indian rupees and exclude taxes. They are one-time setup, not a subscription.
No monthly maintenance fee. No vendor lock-in. The client owns the code and the data.
Talk to salesUsage, such as LLM tokens, is passed through at provider cost. We add no markup.
Bring one call, thread or workflow. We scope from there.
Book a callIntegrations
The brain reads the places your knowledge already lives, keeps each file’s permissions, and answers inside the tool your staff already open.
If your system lets data in or out, we connect to it. If it does not, we find a workaround, and we tell you how dependable that route is before we build. We also connect scanned PDFs, old file shares, intranet pages, email archives and any internal system with an export.
The clean route. Most modern tools have one, and many older ones do too.
No API? We read the database, or a scheduled CSV or SFTP export.
We pick up the reports, invoices and attachments your system already sends.
For a portal with no other way in, a bot works the screen the way your team does, with every step logged.
A company brain reads the files your people already keep: offer letters, client contracts, runbooks, expense policies. Yours may be English documents with Hindi and Kannada forms alongside, spread across Google Drive, Confluence, SharePoint and email. The audit lists which of those sources the brain may read, and which it may not, before anything gets indexed.
Those files hold personal data, so India’s data protection rules shape the design. The Digital Personal Data Protection Rules, 2025 require a Data Fiduciary to inform the Data Protection Board of a breach without delay and to give details within seventy-two hours. That duty applies from 13 May 2027. The PIB backgrounder on the Rules puts penalties at up to ₹250 crore for failing to keep reasonable security safeguards, and says access and correction requests have to be answered within ninety days. A brain helps with the last one: it can list every document that mentions a person. It also needs its own safeguards, so each question and answer is logged with the permission that allowed it.
Where the index sits is a client decision. For most firms we place it in an Indian region of your own cloud account. The Reserve Bank of India requires payment system providers to store payment data only in India, so a fintech client gets that constraint written into the scope. The model that drafts each answer receives the question and the retrieved passages, and we name that provider in the build.
Under Rule 7(2) of the Digital Personal Data Protection Rules, 2025, a Data Fiduciary must tell the Data Protection Board of a personal data breach without delay and give detailed information within seventy-two hours. The rule applies from 13 May 2027.
Gazette of India, G.S.R. 846(E), 13 November 2025Penalties under the DPDP Act run up to ₹250 crore for failing to maintain reasonable security safeguards and up to ₹200 crore for failing to notify a breach. Access and correction requests must be answered within ninety days.
Press Information Bureau, backgrounder on the DPDP Rules, 2025, 17 November 2025The Reserve Bank of India requires payment system providers to store the entire data relating to payment systems only in India, though the foreign leg of a transaction may also be stored abroad.
Reserve Bank of India circular RBI/2017-18/153, 6 April 201817th Cross Rd, Sector 3, HSR Layout, Bengaluru, Karnataka 560102, India
These are published patterns. They are not a claim that the client is in a particular city.
Case study, Infrastructure project operatorAn infrastructure project operator cut answers on project risk from 4 hours to 9 minutesField reports turned into current answers that link back to the passage behind them. Published work on the pattern, with no claim about where the client sits.
EssayEight controls every production AI agent needsAn internal agent still stops at the decision a person owns.That depends on your sector. The RBI requires payment system providers to store payment data only in India. For other firms we place the index in the region your counsel picks, and an Indian region is available from the major cloud providers. We name the model provider that sees each question.
From 13 May 2027 a Data Fiduciary must give the Data Protection Board details of a personal data breach within seventy-two hours. The brain keeps a log of what each query read, which gives you evidence to work from. Your own incident process still owns the notification.
We test retrieval on your documents in each language during the audit. English works best today. If Kannada or Hindi retrieval is weak on your files, we report it with examples and scope the sources accordingly.
Setup is typically from about ₹1,25,000, one-time, scoped after the audit. There is no monthly maintenance fee. Model usage is billed at provider cost.
The first use case is live within a week, and we build on top of it, one use case at a time. We do not quote a total timeline. Which use case comes next, and how many sources and permission rules it involves, is agreed as we go.
Tell us the call, the thread or the record. Every setup on this page starts with a call.
Book a call