# RAG and document AI development in the UAE

> RAG and document AI in the UAE: assistants that answer from your contracts, BOQs and manuals with the source shown, measured before launch. Book a call.

We build assistants and document AI that answer from your contracts, BOQs, manuals, price lists and policies, show the source of each answer, and hand anything uncertain to a person. Accuracy is measured on your own material before you accept the work.

**Service type:** RAG and Document AI Development

## What a RAG or document AI build includes

Every build is measured the same way, whatever documents it reads.

- Search and answers over contracts, BOQs, manuals, policies and price lists
- The source document and passage shown with every answer
- Fields read from documents and written straight into your systems
- A recorded test set from your own material, built before a model is chosen
- Access controls so each person sees only the documents they may see

## Why teams choose innopalm for document AI

- We measure a general model first and fine tune only if it falls short, then re-measure
- Anything out of scope or below the agreed confidence threshold goes to a person
- Values that are not in your master data are refused, never invented
- Code, tests and documentation you own at handover

## How a RAG or document AI project runs

The test set comes first, so the accuracy figure means something.

- **Discovery and scoping** We confirm there is a clear use case and that no off-the-shelf product already does the job, and end with a written scope you approve.
- **The test set** The first two weeks go to collecting 100 to 300 of your real documents and questions into a recorded test set.
- **Build and measure** You see running software every 2 to 3 weeks. We measure a general model on the test set, fine tune only if it falls short, and show you both figures.
- **Accept and keep measuring** You accept the work against the measured accuracy, and the test set is re-run on every model or prompt change. The first release goes live 6 to 12 weeks from kickoff.

- [See AI and machine learning development](https://www.innopalm.com/services/ai-machine-learning-development)
- [See AI agent development for agents that act on documents](https://www.innopalm.com/services/ai-agent-development)

## What a RAG or document AI project costs

The same project floor and payment terms as any build.

- **From AED 50,000, discovery included** The work starts from our AED 50,000 floor with discovery included and is priced above it by the custom AI work: building the test set and measuring accuracy.
- **Milestone payments on acceptance** There is no hourly rate. The figure is fixed once you approve the written scope, and each milestone is paid on acceptance.

## Frequently asked questions

### What is RAG, and when does a business need it?

Retrieval-augmented generation means the model answers from your own documents rather than from memory, and shows where the answer came from. A business needs it when people spend time searching contracts, manuals or past quotes for answers that already exist.

### RAG or fine tuning: which do we need?

We measure a general model with retrieval on your test set first and fine tune only if it falls short, then re-measure on the same set. You see both figures before you accept the work.

### What happens when the assistant is not sure?

It does not guess. Questions that are out of scope or below the confidence threshold we agree with you go to a person on your team.

### How long does a document AI project take?

The first release goes live 6 to 12 weeks from kickoff, with the first two weeks spent building the test set from your real material.

### When do you advise against it?

When there is no clear use case, when an existing off-the-shelf product already does the job, or when the budget is below our AED 50,000 floor.

---

Source: https://www.innopalm.com/services/rag-document-ai | Markdown: https://www.innopalm.com/services/rag-document-ai.md | Site index: https://www.innopalm.com/llms.txt | API: https://www.innopalm.com/openapi.json