Ask a question. It answers from your own documents, and shows where.
Your documents
- Employee handbook.pdf
- Returns policy.docx
- Line 3 machine manual.pdf
services / ai consultancy
AI that answers from your company's own documents.
We build custom AI on your own data: an internal assistant your team asks in plain language, that answers from your files and shows where each answer came from. Scoped, built and handed over running, working directly with the founders.
- the service
- Custom AI on a company's own data
- built by
- The two founders, directly
- in build now
- A research assistant for Niveshaay
> 01 / try it
Ask it something. Watch where the answer comes from.
Pick a question. The answer is written from one of the three documents shown, and that document lights up. An answer your team can check is the whole difference.
A demo on sample data: three documents from an invented company, with fixed answers. Nothing is sent to a model. The real thing is built on your files.
> 02 / what you get
A working system, not a strategy deck.
- 01
A written scope first
Before any build: which questions the system has to answer, which documents and data it reads, who can use it, and what it must never do.
- 02
Answers from your files
An assistant that reads the documents you already have, such as PDF and Word files, and answers your team in plain language.
- 03
A source on every answer
Each answer says which document it came from, so a person can check it in seconds instead of taking it on trust.
- 04
Your rules, built in
The limits you set are part of the system, not a line in a policy. For example: the AI never acts on its own, and nothing is ever deleted.
- 05
A sign-in for your team
It sits behind a sign-in, so the people who can ask it questions are the people you chose.
- 06
Handed over running
Deployed and in use, with a handover so your team knows how it works and how to ask for changes.
> 03 / proof from real work
see all the work →Where we are doing this now.
in build
Niveshaay
A SEBI-registered research and fund-management firm
We are currently building NiveshAI for Niveshaay: an internal web app with two modules behind one sign-in. It is in active daily development and is not live yet.
Read what is being built→- 2
- modules behind one sign-in: Chat and Tasks
- 4
- separate environments: development, demo, test, production
- 0
- actions the AI takes on its own
- Chat: a research assistant that answers from the firm's own files and a market data feed
- Every figure in an answer is cited to its source
- Reads uploaded PDF and Word documents
- Tasks: team task management delivered through Microsoft Teams
- Two rules built in: the AI never acts alone, and nothing is deleted
- Nightly backup
live
The Outbound Engine's reply brain
Our own product, in production
The same idea inside a product we build and run ourselves. An AI reads every inbound reply, classifies it, and drafts an answer grounded in that client's own knowledge base. A person approves it before anything is sent.
- 9
- classifications an inbound reply can be given
- 1
- separate database for each client
- 15 min
- between checks of its own records against outside systems
- Drafts are grounded in each client's own knowledge base
- A person approves every draft, in the dashboard or from a Slack card
- Unknown values are shown as unknown, never as a made-up zero
- An operator console with an append-only audit log
> 04 / how it runs
From a question to a running system.
- 01 /
Scope
A call with the founders. We work out which questions matter, where the answers live today, and whether custom AI is the right tool for them at all.
- 02 /
Build on your documents
We build against your real files, not sample ones, so what you see in progress is what you will get.
- 03 /
Test with your questions
Your team asks it the things they actually need to know. We check answers against their sources and fix what it gets wrong.
- 04 /
Hand over running
It goes live behind your sign-in, with a handover. You keep working with the same two people who built it.
> 05 / frequently asked
Asked before the first call
01What does AI consultancy from mkdir actually include?
Building, not advising. We build custom AI on your company's own data. A typical example is an internal assistant that answers your team's questions from your own documents and shows which document each answer came from. We scope it with you, build it, and hand it over running.
02How is this different from giving the team a general chatbot?
A general chatbot answers from what it learned on the public internet, and it knows nothing about your handbook, your contracts or your research. A system built on your data answers from your files, says where each answer came from, and follows the rules you set. That turns an answer into something your team can check.
03Will it make things up?
Any language model can be wrong, so we build for checking instead of asking you to trust it. Answers are tied to a source document. In the assistant we are building for Niveshaay, every figure is cited to where it came from. When the documents do not contain an answer, the right behaviour is to say so, and we test for that with your own questions.
04What happens to our documents and data?
That is decided before anything is built. Which documents the system reads, where it runs and who can sign in are all written into the scope, and you agree to them first. As one example of how we separate things: in our own Outbound Engine, each client has its own separate database.
05Have you built this before?
Yes, and here is the honest status. We are currently building NiveshAI for Niveshaay, a SEBI-registered research and fund-management firm: a research assistant that answers from the firm's own files with every figure cited. It is in build, not live. In the Outbound Engine, our own product, which is in production, an AI drafts replies grounded in each client's own knowledge base and a person approves every one.
06What does it cost, and how long does it take?
We quote a fixed price up front, based on the scope, and you know the timeline before any work starts. The first call is free, and it ends with a plain answer about whether this is worth building.
$ mkdir build
Have documents your team keeps digging through?
Book a call with the founders. We will tell you honestly whether custom AI is worth building for your case, and what we would build first.