Private AI on your documents
Assistants that answer from your documents, and show their sources.
- from ₹50,000two-week sprint, ends in a fixed quote
We build retrieval-augmented (RAG) assistants that answer questions from your policies, manuals, contracts and knowledge base, citing the exact source for every answer. They run on hosted models such as Claude and OpenAI, or on self-hosted open models when your data must stay on your own servers.
Right for you if
Who this is for.
- Support and operations teams answering the same questions from long documents
- Regulated businesses that cannot send data to a public chatbot
- Product teams adding search or Q&A on top of their own content
Each assistant ships with an evaluation set so you can see how accurate it is.
What you get
What the assistant comes with.
Document pipeline
Ingestion, cleaning and chunking of PDFs, docs, wikis and databases, kept in sync.
Retrieval and citations
Every answer links to the passage it came from, so people can check it.
Hosted or self-hosted
Claude or OpenAI, or open models on your infrastructure when data residency matters.
Evaluations
A test set of real questions scored for accuracy and refusal behaviour, rerun on every change.
Access control
People only get answers from documents they are allowed to see.
Interfaces
Web, Slack, WhatsApp or an API, depending on where your team works.
Accuracy
How you can check the answers are right.
Every assistant ships with a test set built from your real questions. This is the format; the numbers come from your documents.
| Real question | Expected source | Result |
|---|---|---|
| What is the notice period for staff on probation? | HR policy, section 4 | pass / fail |
| Can a customer return an opened item? | Returns policy, page 2 | pass / fail |
| Question with no answer in the documents | None: should say it does not know | pass / fail |
Illustration of the format, not a client result. You see the score before launch and after every change.
Examples
Questions it can answer for your team.
Illustrations of what we build, not client claims.
- An internal policy assistant for HR and operations questions
- A product-documentation assistant for support agents and customers
- Contract and tender search with clause-level citations
How it works
Every project starts with a two-week sprint.
Sprint · 2 weeks
Interviews, a clickable prototype, scope, architecture and a fixed quote. From ₹50,000, credited to the build.
Build
Weekly releases you can try. Fixed scope, milestone billing.
Launch
Production release, monitoring and launch fixes.
Run (optional)
Retainer or care plan to run and improve it.
Built with
- Claude
- OpenAI
- Open models (self-hosted)
- PostgreSQL + pgvector
- Python
- Node.js
- Evaluation harnesses
Questions
Private AI on your documents: straight answers.
What is RAG?
Retrieval-augmented generation: the assistant first finds the relevant passages in your documents, then answers using only those passages and cites them. It is how you get answers grounded in your own content instead of the model's general knowledge.
Can the AI run on our own servers?
Yes. When data must not leave your infrastructure we deploy open models on your servers or private cloud. We will tell you honestly where open models trade accuracy for control.
How do we know the answers are right?
Every assistant ships with an evaluation set of real questions and expected answers. You see the score before launch and after every change.
How much does an AI chatbot on our documents cost?
It starts with the two-week sprint from ₹50,000, which ends in a fixed quote for your documents and users. Model usage and hosting are separate running costs, estimated during the sprint.
Can we just upload our documents to ChatGPT?
For one person's own questions, often yes. Build an assistant when many people need answers, when answers must show their source, when some documents are restricted to some people, or when the data cannot go to a public chatbot.
Related
Other things we build.
Written scope, fixed quote
Tell us what slows your business down.
Websites and automation are quoted after a 30-minute call. App and software builds start with a two-week sprint that ends in a written scope and a fixed quote.
- 30-min call
- Written scope
- See the price list →