About — a product of BizfyLabs

An engineering company in Dubai, building document AI that never leaves your building

DocxIntel is built by BizfyLabs, a UAE-based engineering company. We started it for a straightforward reason: the organisations we work with in this region were being offered document AI they were not permitted to use. This page explains who we are, what we believe about the architecture, and how we contract.

Dubai, United Arab Emirates
On-premise first, not cloud-first
Open weights in the bundle
Arabic as a primary script

Dubai
Where the company is based: On-premise
The default deployment: AR + EN
Primary scripts: Unmetered
Pages processed

We publish no customer counts, funding figures or headcounts on this site, because DocxIntel is a new product and inventing them would undermine everything else we ask you to believe. The numbers we do publish — accuracy, throughput, model licences — come with the methodology attached.

The company behind it

What BizfyLabs is

BizfyLabs is an engineering company based in Dubai, United Arab Emirates, working with organisations whose systems carry regulatory weight. DocxIntel is its document intelligence product: the same team designs it, deploys it and supports it, and BizfyLabs FZC LLC is the entity you contract with.

Based in Dubai, working primarily with UAE and wider GCC organisations
Engineering-led: the people who build the platform are the people who deploy it
BizfyLabs FZC LLC holds the licence, the Proof of Value agreement and the support relationship
Arabic and English are both working languages, in the product and in the room
No separate DocxIntel legal entity — one commercial counterparty, one contract

DocxIntel is a product line, not a subsidiary. Everything you sign, from the pilot agreement to the annual licence, is with BizfyLabs.

The engineering thesis

Four decisions we made before writing the first line

These are not features. They are constraints we accepted at the start, and every subsequent design decision had to fit inside them. They are also the reason DocxIntel is a different product from a cloud platform with a self-hosting option.

Decision 1

On-premise first, not on-premise eventually

Most document AI is built cloud-first and grows a self-hosting tier once large customers demand it. That order shows up in the architecture: hosted control planes, licensing check-ins, cloud fallbacks for hard pages. We built for a machine with no route to the internet, so an air-gapped install is the ordinary case and a connected one is the convenience.

No phone-home, no telemetry requirement, no fallback cloud call
Identical function connected or disconnected
Included on every licence rather than gated behind a top tier

Decision 2

Open weights, shipped in the box

The models arrive inside the deployment bundle, as artifacts on storage you control, with a manifest naming each model, its version and its licence. If the network goes away, the deployment keeps working. If the commercial relationship goes away, you still have the artifacts already in your data centre, subject to the licence terms you signed.

Full model manifest handed over on day one
Licensed for on-premise and air-gapped commercial use
No per-inference or royalty model fees on top

Decision 3

Arabic as a primary script, not a language pack

Right-to-left reading order, Arabic-Indic numerals, diacritics, handwritten Arabic and pages that mix Arabic body text with English headers are handled at the layout and reading-order layer, not patched on afterwards. In this region a bilingual page is the normal page, and a system that treats it as an edge case fails on the documents that matter most.

Reading order resolved per block, so bilingual tables stay intact
Arabic-Indic and Western numerals both recognised
Handwritten Arabic treated as content, not noise

Decision 4

Traceable by design, because someone has to sign it

Every extracted value carries the page number, the bounding-box coordinates and the confidence score it came from. That is not a reporting feature; it is what makes an internal auditor, a regulator or a claims manager able to accept an automated answer at all. An answer you cannot trace is an answer you cannot sign off.

Coordinates and confidence on every field, not a document-level average
Low-confidence regions flagged for review rather than silently guessed
Stable element identifiers, so diffs across model versions are reviewable

Why we built it

The organisations we work with could not use the products on offer

This was not a market-sizing exercise. It came out of repeated conversations with banks, insurers, hospitals and government departments in the region who had a genuine document problem, had evaluated the cloud options, and had been told no by their own regulator, legal team or security function.

CBUAE outsourcing and data-residency expectations rule out sending customer files to a foreign endpoint
UAE PDPL and the Health ICT Law put localisation obligations on exactly the documents that most need automating
Legal and investigation material carries confidentiality duties that make third-party transmission a professional problem
Per-page pricing turns a 10-million-page archive into a permanent budget line rather than a one-off project
And the documents themselves — phone photographs, handwritten Arabic, stamps over printed text, multi-generation faxes — are the ones generic tools quietly send to manual review

On-premise deployment does not make an organisation compliant by itself. It removes the cross-border transfer and third-party processing questions from the assessment, which is usually the part that stops the project.

An Arabic claim form with extracted fields traced back to their source coordinates on the page

Instead of asking for trust

What a new product can offer in place of a customer list

We are early, and a reference call is the one thing we cannot manufacture honestly. So the argument has to be made with artifacts you can inspect and tests you can run yourself.

A model manifest, not a black box

Every model in the bundle is named, versioned and licensed in a document you receive on day one. You can check the licences against your own legal position before anything is installed.

See what's in the box

Accuracy with its methodology attached

The 99% field-level figure is published alongside the document mix, the adjudication process and the per-field breakdown, because an accuracy number without a methodology is a marketing number.

Read the benchmark

A data-flow diagram you can test

We publish where data moves inside a deployment. During a pilot you can put that diagram next to your own firewall, DNS and proxy logs and confirm it, which is a stronger claim than any certificate we could show you.

See the security model

A pilot with the threshold agreed first

The Proof of Value fixes the accuracy threshold and the conversion price in writing before it starts. Miss the threshold and you keep the report and owe nothing further. The risk sits with us, in the contract.

How the pilot works

Comparisons that name competitor strengths

Our comparison pages state what the other products are genuinely good at and where we would tell you to buy them instead. A comparison page that finds no competitor strengths is an advertisement.

Read the comparisons

No certifications we do not hold

We describe how the architecture supports your own certification and how controls map to the frameworks you are assessed against. We do not claim badges we have not earned, and you should check that claim too.

How controls map

Reference

How DocxIntel and BizfyLabs relate, commercially

The practical answers procurement asks for in the first call, in one place.

Company and product

Product: DocxIntel — on-premise document intelligence platform
Built by: BizfyLabs, an engineering company based in Dubai, UAE
Contracting entity: BizfyLabs FZC LLC
Relationship: DocxIntel is a product line, not a separate legal entity
Primary market: United Arab Emirates and the wider GCC, with deployments elsewhere on request
Working languages: Arabic and English, in the product and in delivery

How we contract

Entry point: A paid Proof of Value: 30-45 days, fixed fee, deployed in your environment
Pilot terms: Accuracy threshold and conversion price agreed in writing before the pilot begins
If the threshold is missed: You keep the report and owe nothing further
Licence model: Fixed annual licence sized by deployment footprint, never per page
Deployment models: Air-gapped, private cloud, managed single-tenant, or evaluation sandbox
Route to market: Direct, and through implementation, infrastructure and regional partners

Support and continuity

Diagnostics without remote access: Log bundles your team reviews and releases, plus scheduled escorted sessions where deeper access is needed
On-site attendance: Available for air-gapped and classified environments where no remote path exists
Updates: Signed offline packages imported on your schedule, with changelog and re-validation report
Rollback: Previous model versions retained, with a documented rollback path
Security pack: Architecture, control mapping and data-flow documentation available under NDA

Specific contractual terms, service levels and regional arrangements are confirmed in the agreement rather than on a web page. Ask for them in writing during the Proof of Value.

How we work

What working with us in the region actually looks like

A deliberately short path from first conversation to a system your auditors can sign off. No unpaid pilots that drift for six months, and no proof of concept that runs on our documents rather than yours.

Step 1
A technical conversation, not a discovery call
We start with your document mix, your volume, your residency constraints and the systems the output has to land in. If DocxIntel is the wrong answer — and for low-volume, cloud-permitted workloads it often is — we would rather establish that in the first hour.
Step 2
The accuracy threshold is agreed in writing
Before any pilot begins, we agree the document set, the fields that matter, how accuracy will be measured field by field, the threshold that counts as success, and the price you would pay if it is met.
Step 3
Deployment inside your perimeter
Air-gapped, private cloud or managed single-tenant, sized against your real volume. Our engineers work alongside your infrastructure and security teams, and the install reaches out to nothing during or after setup.
Step 4
Measurement on your documents, adjudicated once
Field-level scoring against a ground truth your own people adjudicate. You receive the full report — including the fields we did poorly on — whether the engagement converts or not.
Step 5
Handover, then support built for isolated networks
Runbooks, sizing guidance and the model manifest go to your operations team. Ongoing support assumes no remote access by default, and updates arrive as signed offline packages on your schedule.

Questions about the company behind the product

Who you are contracting with, why the product exists, and what we will not claim.

Who builds DocxIntel?

DocxIntel is a product of BizfyLabs, an engineering company based in Dubai, United Arab Emirates. BizfyLabs FZC LLC holds the commercial relationship: it licenses the platform, runs the Proof of Value, delivers the deployment and provides support. There is no separate DocxIntel legal entity to contract with.

Is DocxIntel a new product?

Why build an on-premise platform when the market went to the cloud?

Because a large number of organisations in this region cannot use metered cloud document AI, whatever its quality. Banking, insurance, healthcare, government and legal buyers work under residency and outsourcing rules that make transmitting a document to a third-party endpoint the problem, not the price. Building cloud-first and adding an enterprise self-hosting tier later produces a different product from building on-premise from the first line.

Do you sell directly, or through partners?

How do you support a deployment you cannot connect to?

What do you publish that we can actually verify?

Proof of ValueThe paid 30-45 day pilot, with the accuracy threshold and conversion price fixed up front.Security architectureHow the deployment supports your own certification, control framework and audit scope.Partner programmeIntegrators, infrastructure vendors, cloud partners and GCC resellers — open and early.Deployment modelsAir-gapped, private cloud, managed single-tenant and evaluation sandbox, in detail.BizfyLabsThe engineering company behind DocxIntel, and the rest of what it builds.CareersEngineering and delivery roles at BizfyLabs, listed on the company site.

Start with a conversation about your documents.

Bring the document mix that currently goes to manual review. We will tell you honestly whether an on-premise platform is the right answer for it, and what a Proof of Value would cost.

No per-page metering
Runs in your environment
Written accuracy threshold