It answers only from your documents. It shows where each answer came from. It says "I don't know" instead of making things up.
No deck. No pitch. Just your documents.
Northstr (northstrdata.com) is an AI consultancy that builds private AI assistants on your company's documents.
Built for teams that run on documents
Northstr does. We build assistants that answer only from your documents, show where each answer came from, and say "I don't know" instead of making things up. It starts with the $750 Hundred Questions test.
The answer is in a manual somewhere. People ask a colleague, guess, or wait. Time goes on searching instead of working.
A chatbot that sounds sure but is wrong is worse than no chatbot. Nobody can tell which answers to trust.
Your policies and product details should not end up in a public tool. You need to know where your documents go.
AI bills come as one monthly number. Nobody knows what one answered question costs, so nobody can plan.
The screens below use sample data. On a call we show you the real thing.
A private assistant on your own content: manuals, policies, training material, product docs. Staff or customers get instant, sourced answers. It runs in accounts you own.
For training companies: learners get sourced answers from your own material, and your staff see what learners ask.
Ask your data questions in plain English and get answers checked against your own definitions. We build it on request.
Here is the list, worked out from your data using your own definition of a return.
✓ Checked against your metric: returnsIt books appointments, takes messages and passes hot leads to a person. It is in early access, so talk to us first.
We work out what one task costs, such as one answered question, instead of looking at the monthly bill. Then you can plan.
Get the free check →One task, one cost. The monthly bill is one big number. We break it down to the cost of a single answered question, so you know if it makes sense to grow.
Four steps. You see how it works on your own documents before you commit to more.
The Hundred Questions test. We build a working assistant on a sample of your documents. Your team sends 100 real questions, and a person checks every answer. 1 week.
If the test goes well, we build the final version on all your documents. About 3 more weeks.
It runs in accounts you own. You get the keys, the setup and a walk-through for your team.
Optional monthly support. We reply within 1 US business day and update it when your documents change.
An education company in a regulated industry had an AI tutor that kept making up answers. More rules did not stop it. We rebuilt it around one rule: the AI only explains, and only from the company's learning material. Simple software asks the questions and grades the answers against the real answer key. The AI never decides right or wrong. Before each release we run 150 test messages written the way learners really type, and a person reads every reply.
Read the full case study →We built AI call scoring for a US financial services company. It costs about 2.5¢ in AI fees to transcribe and score a call. Call scoring is now Callyst →
The AI Assistant starts with one small, fixed-price step. You see results on your own documents before you commit to more.
A private assistant that answers only from your own documents.
A working version on a sample of your documents. 100 real questions from your team, each answer checked by a person. A report of right, wrong and "I don't know" answers.
1 week $750 fixed What you get →The final version on all your documents, handed over to you. Your fixed build price is in your Hundred Questions report.
About 3 more weeks From $3,000We reply within 1 US business day and update it when your documents change.
Monthly, optional Price given on the callWho pays for the AI? Your AI running costs (OpenRouter, AWS) are billed to your own accounts. Our fee covers building the system, and we never mark up your AI bill.
It books appointments, takes messages and passes hot leads to a person.
Book a call →We work out what one task costs, such as one answered question, instead of looking at the monthly bill.
Get the free check →10+ years in data and AI across fintech, healthcare, e-commerce and gaming. Built and runs AI call scoring for a US financial services company.
AI researcher with a software engineering background. Runs model evaluation at Northstr. Every model we ship is benchmarked against your real data before it touches production.
We build systems that run in production. You are not our test case.
Cheaper, auditable, and not locked to one vendor. We cut AI bills, not corners.
A 1-week test, then a 3-week build.
Your accounts, your keys, your data. We never mark up your AI bill.
Northstr does. We build assistants that answer only from your documents, show where each answer came from, and say "I don't know" instead of making things up. It starts with a $750 Hundred Questions test.
Teams that run on documents: policies, manuals, training material, product docs and support. We are based in Delhi NCR and work with US companies.
Hundred Questions: $750, 1 week. Full build: from $3,000, about 3 more weeks. Your fixed build price is in your Hundred Questions report. Monthly support is optional, and we give you the price on the call. Your AI running costs (OpenRouter, AWS) are billed to your own accounts, and we never mark them up.
A sample of your documents, 100 real questions from your team, and one contact person. The week starts the day we have them. A sample report is available on request by email.
The assistant may only answer from your documents. Every answer shows its sources. When the documents do not have the answer, it says "I don't know". In the test, a person checks every answer. Read how to stop an AI assistant making up answers.
Your documents are stored in accounts you own. Before we start, we tell you exactly which AI providers see your data.
The Hundred Questions test takes 1 week. The full build takes about 3 more weeks.
Yes. Call scoring is now its own product, Callyst. See callyst.com.
We prefer it. We embed alongside your people and transfer knowledge, or act as your full fractional AI team if you do not have one in-house.