Client Support
Contact
EN / FR
Products
Services
Solutions
Our Work
Resources
About
GET STARTED →
[ AI & MACHINE LEARNING ]
AI development services,
from first use case
to production.
Dirox designs, builds, and runs AI applications for companies that need AI working in real products and operations: assistants that answer from your own data, models that classify and predict, and automation that removes manual steps. We first check that AI is the right answer. Our engineers then build it to production standards and keep it running.
Fixed price from US$X · 3 weeks · A report and roadmap you keep
Engineering in Ho Chi Minh City, Vietnam, since 2003.
Your data stays in the region you choose.
You own the code, the prompts, and the models we train for you.
[ PROTOTYPE TO PRODUCTION ]
Your AI works. Now make it a product.
Many teams now have a working AI prototype, often built quickly with AI coding tools. It demos well, but it isn't ready for paying customers. We close that gap and keep what already works.
TESTS
Automated tests, evaluation sets that check the AI's answers, and releases with review and rollback.
SECURITY
Authentication, secrets management, dependency checks, and a review of the code your tools generated.
MULTI-TENANCY
Separate data and permissions for each customer, with tests that check the separation.
HOSTING
Production cloud infrastructure in the region you choose, with monitoring and backups.
COST CONTROL
Token budgets, caching, and model routing, so your AI bill stays predictable as usage grows.
GOVERNANCE
Logs, human review where it matters, and the documentation an enterprise buyer will ask for.
BOOK A PROTOTYPE‑TO‑PRODUCTION ASSESSMENT →
2 weeks · from US$X
The infrastructure half of this story is on Cloud & DevOps. The full path is on Turn a Prototype into a Product.
[ WHAT WE DO ]
Three ways we put AI to work.
Our AI work falls into 3 areas, from advice to finished products.
AI CONSULTING & IMPLEMENTATION
USE-CASE SELECTION
Find the AI use cases worth funding, estimate what each will cost to build and to run, and decide what to do first.
AI INTEGRATION
Add AI to the software you already run, without rebuilding it: search, summaries, document understanding, and assistants inside existing screens.
MODEL CHOICE AND GOVERNANCE
Compare hosted and open-source models on your own test cases, set up human review, and document each decision for GDPR and the EU AI Act.
AI PRODUCT DEVELOPMENT
AI ASSISTANTS AND AGENTS
Assistants that answer from your documents and data, and agents that take defined actions in your systems, within limits you set.
MACHINE LEARNING MODELS
Recommendation, classification, and prediction models trained on your data, with accuracy you can measure.
GENERATIVE AI PRODUCTS
Products built around generated text, images, or video, with the review steps and usage limits a business needs.
BUSINESS PROCESS AUTOMATION
DOCUMENT PROCESSING
Read invoices, forms, and contracts with AI document understanding, which copes with layouts that fixed templates miss, and send the results where they belong.
DATA COLLECTION AND ENRICHMENT
Collect, clean, and structure data from many sources, such as supplier catalogs and public websites, and keep it current.
WORKFLOW AUTOMATION
Connect AI steps to your CRM, ERP, and internal tools, so routine work moves without copy and paste.
[ FAST DECISION MODELS ]
Not every decision needs a large language model.
Many business decisions are narrow and repeated thousands of times: route this ticket, tag this document, score this lead, flag this transaction. For these, we use small decision models, sometimes called System 1 AI. They pick from a fixed set of answers you define and report how confident they are.
Because they only choose among your options, they don't invent answers. They can still be wrong, so we measure their accuracy on your own examples and send low-confidence cases to a person. Open-source versions can run inside your own environment.
[ OUR APPROACH ]
Find the right use for AI.
Every Dirox engagement follows the same 3 phases: Insights tells you what is worth building and what it will cost, Studio builds it, and Shield keeps it secure and improving.
01
IDENTIFY · INSIGHTS
Start with the use case.
We define the problem, the users, the value, and how you will measure success.
02
EXPLORE · INSIGHTS
Check what is possible.
We review your data, systems, and model options, and estimate the build cost and the monthly running cost.
03
BUILD · STUDIO
Build it to production standards.
We build at a fixed price on the scope agreed in Insights, with tests, security reviews, and evaluation sets for the AI's answers.
04
INTEGRATE · STUDIO
Put it where the work happens.
We connect the AI to the applications and workflows your teams already use, with access rules and logs.
05
IMPROVE · SHIELD
Keep it accurate and affordable.
Identify and explore in Insights. Build and integrate in Studio. Improve with Shield.
[ OUR WORK ]
AI built around real business needs.
A few of the AI systems we have delivered. Each case study sets out the problem, what we built, and what changed.
[ WAYS TO START ]
Start with a fixed price and a clear deliverable.
Pick the starting point that matches where you are. Each one has a fixed price, a set duration, and a result you keep, whether or not you continue with us.
[ DATA, SECURITY, AND COMPLIANCE ]
Your data stays yours.
Before you share any data, you should know where it goes, who can see it, and whether anyone trains on it. These are the rules we work by.
Your data:
Your data stays in the region you choose, normally your home country.
We build on Google's Vertex AI, in a Google Cloud account you own, where your data is protected by your enterprise agreement, customer-managed encryption keys (CMEK), and data residency.
Your data is never used to train models for anyone else, and we never sell or trade it. Models we train on your data are yours.
If you prefer, we run open-source models on your own servers or in your own cloud account.
Your access:
You own the cloud accounts and the keys.
Our access is individual, logged, and revocable by you alone.
We work without production data by default.
Everything we build is documented so another team could take over.
How we use AI in our own delivery:
Engineers review every AI-assisted change, and each change must pass unit tests, static security analysis, and secret scanning before release.
We use enterprise licenses for our AI tools, so your code and data are not used for training.
The rules you work under:
We help you meet GDPR, the EU AI Act, NIS2, DORA, and Singapore's PDPA. For each AI feature, we document how it handles personal data and where it sits in the EU AI Act's risk categories, for your legal team to review.
Partnerships:
We hold no commercial partnership with AI model providers today. If we sign one that could influence what we recommend, we will say so here and in your assessment report.
We operate to security policies aligned with ISO 27001. We are working toward a Cyber Trust Mark (CTM) Certification.
READ OUR TRUST & SECURITY POLICIES →
[ WHY DIROX ]
AI needs more than a good model.
The model is a small part of a working AI system. The rest is engineering, data, security, and the people who will use it every day.
01
VIETNAMESE ENGINEERING, EUROPEAN GOVERNANCE
Engineers in Ho Chi Minh City, rules you can review.
We have built software in Ho Chi Minh City, Vietnam, since 2003. You keep the keys, our access is logged, every handover is documented, and our security policies are aligned with ISO 27001.
02
COMPLIANCE-AWARE AI
Privacy and risk are planned before we build.
We map each AI feature to the rules that apply to it, such as GDPR and the EU AI Act, before we build it. When something is unclear, we say so and come back to you in writing.
03
AI-ASSISTED DELIVERY, ENGINEERS IN CONTROL
Faster delivery at a locked price.
We use AI tools in our own work to move faster on the same scope. You get that speed as an earlier delivery date at a fixed price.
04
BUILD AND GO-TO-MARKET TOGETHER
We help you sell what we build.
Your digital Sherpa for AI: we know the route, and we stay for the whole climb.
[ EXPLORE FURTHER ]
What a working AI feature depends on.
Data, software, infrastructure, and security all shape how well your AI works.
[ QUESTIONS ]
Questions to ask
before you build AI.
[ ASK NIMA ]
Not sure where to start?
Nima, our assistant, answers from the pages of this site and passes your question to an expert when you want one.
ASK NIMA →
[ WHERE ARE YOU HEADING? ]
Tell us what you want AI to do for you.
Book an assessment, talk to an expert, or ask Nima, our assistant, a first question.
BOOK AN AI USE‑CASE ASSESSMENT →
3 weeks · from US$X
TALK TO AN EXPERT →
ASK NIMA →
Dirox is part of the Synexser group. Engineering in Ho Chi Minh City, Vietnam, since 2003.
PRODUCTS
Insights
Studio
Shield
SERVICES
AI & Machine Learning
Data Management & Governance
Odoo & ERP
Cloud & DevOps
Cybersecurity & Compliance
Software Architecture & Modernization
Software Development
Maintenance & Managed Services
Go-to-Market
SOLUTIONS
Agritech
Construction
Financial Services & Insurance
Healthcare & Life Sciences
Logistics & Supply Chain
Manufacturing
Retail & Consumer Goods
All solutions
RESOURCES
Our Work
Blog
Events & Webinars
News
Guides & Downloads
ABOUT
Our Story
Why Dirox
Our Team
Trust & Security
Partners
Dirox Gives
Careers
Locations & Contact
OFFICES
[Headquarters: city, registered address]
[Office: city, registered address]
[Office: city, registered address]
© 2026 Dirox. All rights reserved.
Legal Notice
Privacy Policy
Cookies Policy
Cookie settings
Client Support
Where does our data go, and do you train on it?
Where does our data go, and do you train on it?
Your data stays in the cloud region you choose, normally your home country. We build on Google's Vertex AI in a cloud account you own, where your data is covered by your enterprise agreement, customer-managed encryption keys, and data residency, or on open-source models in your own environment. Your data is never used to train models for anyone else, and we never sell or trade it.
Who owns the models, prompts, and code, and can we take them elsewhere?
Who owns the models, prompts, and code, and can we take them elsewhere?
You do, and you can. You own all the intellectual property we produce for you or with you: code, prompts, evaluation sets, the models we train for you, and documentation. Dirox keeps only its own tools, methods, and know-how. Third-party models stay under their providers' licenses, and we show you those terms before you commit. Everything is documented, so another team can take over.
Which models do you use, and how do we avoid lock-in?
Which models do you use, and how do we avoid lock-in?
We choose per use case, after testing on your own examples: hosted models such as Gemini on Vertex AI, or open-source models we run on your infrastructure. We put a thin layer between your application and the model, so you can switch providers without rewriting the product, and we keep the test sets that prove a new model is as good as the old one.
What will it cost to run once it's live?
What will it cost to run once it's live?
Running costs come mainly from inference (tokens or GPU time), hosting, and monitoring. We estimate them for each use case before you build, during the AI Use-Case Assessment. We keep them predictable with caching, smaller models where they are good enough, decision models, and usage limits. Once it's live, Shield Advanced or Premier reports your actual usage and cost every month.
How long does a proof of concept take, and what does it cost?
How long does a proof of concept take, and what does it cost?
Most proofs of concept run as a 2 to 4-week Studio Sprint at a fixed price, from US$X, agreed after the assessment. You see the AI working on your own data, with its accuracy measured on your own examples, before you decide on a full build. If the results don't justify one, you stop there and keep what we built.
Who works on our project, and do you use subcontractors?
Who works on our project, and do you use subcontractors?
A lead who speaks your language, English or French, and engineers in Ho Chi Minh City, Vietnam (UTC+7), all Dirox employees. We send a written update every week. If a project needs a specialist subcontractor, we name them in the proposal before work starts.
Once it's live, what support hours and response times apply?
Once it's live, what support hours and response times apply?
Every system we deliver includes Shield Essential: for 12 months from go-live, we respond to product defects within 24 hours on business days and fix them at no charge. Shield Advanced adds monitoring, security updates, version upgrades, and support with no cap on hours, and responds within 12 hours, 24/7/365. Shield Premier responds to a critical incident within 2 hours. For AI systems, Advanced also tracks token costs and checks answer quality every month.
GET STARTED →
[ AI & MACHINE LEARNING ]
AI development
services, from
first use case
to production.
Dirox designs, builds, and runs AI applications for companies that need AI working in real products and operations: assistants that answer from your own data, models that classify and predict, and automation that removes manual steps. We first check that AI is the right answer. Our engineers then build it to production standards and keep it running.
Fixed price from US$X · 3 weeks · A report and roadmap you keep
Engineering in Ho Chi Minh City, Vietnam, since 2003.
Your data stays in the region you choose.
You own the code, the prompts, and the models we train for you.
[ PROTOTYPE TO PRODUCTION ]
Your AI works.
Now make it a product.
Many teams now have a working AI prototype, often built quickly with AI coding tools. It demos well, but it isn't ready for paying customers. We close that gap and keep what already works.
TESTS
Automated tests, evaluation sets that check the AI's answers, and releases with review and rollback.
SECURITY
Authentication, secrets management, dependency checks, and a review of the code your tools generated.
MULTI-TENANCY
Separate data and permissions for each customer, with tests that check the separation.
HOSTING
Production cloud infrastructure in the region you choose, with monitoring and backups.
COST CONTROL
Token budgets, caching, and model routing, so your AI bill stays predictable as usage grows.
GOVERNANCE
Logs, human review where it matters, and the documentation an enterprise buyer will ask for.
BOOK A PROTOTYPE‑TO‑PRODUCTION ASSESSMENT →
2 weeks · from US$X
The infrastructure half of this story is on Cloud & DevOps. The full path is on Turn a Prototype into a Product.
[ WHAT WE DO ]
Three ways we
put AI to work.
Our AI work falls into 3 areas, from advice to finished products.
AI CONSULTING & IMPLEMENTATION
USE-CASE SELECTION
Find the AI use cases worth funding, estimate what each will cost to build and to run, and decide what to do first.
AI INTEGRATION
Add AI to the software you already run, without rebuilding it: search, summaries, document understanding, and assistants inside existing screens.
MODEL CHOICE AND GOVERNANCE
Compare hosted and open-source models on your own test cases, set up human review, and document each decision for GDPR and the EU AI Act.
AI PRODUCT DEVELOPMENT
AI ASSISTANTS AND AGENTS
Assistants that answer from your documents and data, and agents that take defined actions in your systems, within limits you set.
MACHINE LEARNING MODELS
Recommendation, classification, and prediction models trained on your data, with accuracy you can measure.
GENERATIVE AI PRODUCTS
Products built around generated text, images, or video, with the review steps and usage limits a business needs.
BUSINESS PROCESS AUTOMATION
DOCUMENT PROCESSING
Read invoices, forms, and contracts with AI document understanding, which copes with layouts that fixed templates miss, and send the results where they belong.
DATA COLLECTION AND ENRICHMENT
Collect, clean, and structure data from many sources, such as supplier catalogs and public websites, and keep it current.
WORKFLOW AUTOMATION
Connect AI steps to your CRM, ERP, and internal tools, so routine work moves without copy and paste.
[ FAST DECISION MODELS ]
Not every decision needs a large language model.
Many business decisions are narrow and repeated thousands of times: route this ticket, tag this document, score this lead, flag this transaction. For these, we use small decision models, sometimes called System 1 AI. They pick from a fixed set of answers you define and report how confident they are.
Because they only choose among your options, they don't invent answers. They can still be wrong, so we measure their accuracy on your own examples and send low-confidence cases to a person. Open-source versions can run inside your own environment.
[ OUR APPROACH ]
Find the right use for AI.
Every Dirox engagement follows the same 3 phases: Insights tells you what is worth building and what it will cost, Studio builds it, and Shield keeps it secure and improving.
01
IDENTIFY · INSIGHTS
Start with the use case.
We define the problem, the users, the value, and how you will measure success.
02
EXPLORE · INSIGHTS
Check what is possible.
We review your data, systems, and model options, and estimate the build cost and the monthly running cost.
03
BUILD · STUDIO
Build it to production standards.
We build at a fixed price on the scope agreed in Insights, with tests, security reviews, and evaluation sets for the AI's answers.
04
INTEGRATE · STUDIO
Put it where the work happens.
We connect the AI to the applications and workflows your teams already use, with access rules and logs.
05
IMPROVE · SHIELD
Keep it accurate and affordable.
Identify and explore in Insights. Build and integrate in Studio. Improve with Shield.
[ OUR WORK ]
AI built around
real business needs.
A few of the AI systems we have delivered. Each case study sets out the problem, what we built, and what changed.
[ WAYS TO START ]
Start with a fixed price
and a clear deliverable.
Pick the starting point that matches where you are. Each one has a fixed price, a set duration, and a result you keep, whether or not you continue with us.
[ DATA, SECURITY, AND COMPLIANCE ]
Your data stays yours.
Before you share any data, you should know where it goes, who can see it, and whether anyone trains on it. These are the rules we work by.
Your data:
Your data stays in the region you choose, normally your home country.
We build on Google's Vertex AI, in a Google Cloud account you own, where your data is protected by your enterprise agreement, customer-managed encryption keys (CMEK), and data residency.
Your data is never used to train models for anyone else, and we never sell or trade it. Models we train on your data are yours.
If you prefer, we run open-source models on your own servers or in your own cloud account.
Your access:
You own the cloud accounts and the keys.
Our access is individual, logged, and revocable by you alone.
We work without production data by default.
Everything we build is documented so another team could take over.
How we use AI in our own delivery:
Engineers review every AI-assisted change, and each change must pass unit tests, static security analysis, and secret scanning before release.
We use enterprise licenses for our AI tools, so your code and data are not used for training.
The rules you work under:
We help you meet GDPR, the EU AI Act, NIS2, DORA, and Singapore's PDPA. For each AI feature, we document how it handles personal data and where it sits in the EU AI Act's risk categories, for your legal team to review.
Partnerships:
We hold no commercial partnership with AI model providers today. If we sign one that could influence what we recommend, we will say so here and in your assessment report.
We operate to security policies aligned with ISO 27001. We are working toward a Cyber Trust Mark (CTM) Certification.
READ OUR TRUST & SECURITY POLICIES →
[ WHY DIROX ]
AI needs more
than a good model.
The model is a small part of a working AI system. The rest is engineering, data, security, and the people who will use it every day.
01
VIETNAMESE ENGINEERING, EUROPEAN GOVERNANCE
Engineers in Ho Chi Minh City, rules you can review.
We have built software in Ho Chi Minh City, Vietnam, since 2003. You keep the keys, our access is logged, every handover is documented, and our security policies are aligned with ISO 27001.
02
COMPLIANCE-AWARE AI
Privacy and risk are planned before we build.
We map each AI feature to the rules that apply to it, such as GDPR and the EU AI Act, before we build it. When something is unclear, we say so and come back to you in writing.
03
AI-ASSISTED DELIVERY, ENGINEERS IN CONTROL
Faster delivery at a locked price.
We use AI tools in our own work to move faster on the same scope. You get that speed as an earlier delivery date at a fixed price.
04
BUILD AND GO-TO-MARKET TOGETHER
We help you sell what we build.
Your digital Sherpa for AI: we know the route, and we stay for the whole climb.
[ EXPLORE FURTHER ]
What a working
AI feature depends on.
Data, software, infrastructure, and security all shape how well your AI works.
[ QUESTIONS ]
Questions to ask
before you build AI.
[ ASK NIMA ]
Not sure where to start?
Nima, our assistant, answers from the pages of this site and passes your question to an expert when you want one.
ASK NIMA →
[ WHERE ARE YOU HEADING? ]
Tell us what you
want AI to do for you.
Book an assessment, talk to an expert, or ask Nima, our assistant, a first question.
BOOK AN AI USE‑CASE ASSESSMENT →
3 weeks · from US$X
TALK TO AN EXPERT →
ASK NIMA →
Dirox is part of the Synexser group. Engineering in Ho Chi Minh City, Vietnam, since 2003.
PRODUCTS
Insights
Studio
Shield
SERVICES
AI & Machine Learning
Data Management & Governance
Odoo & ERP
Cloud & DevOps
Cybersecurity & Compliance
Software Architecture & Modernization
Software Development
Maintenance & Managed Services
Go-to-Market
SOLUTIONS
Agritech
Construction
Financial Services & Insurance
Healthcare & Life Sciences
Logistics & Supply Chain
Manufacturing
Retail & Consumer Goods
All solutions
RESOURCES
Our Work
Blog
Events & Webinars
News
Guides & Downloads
ABOUT
Our Story
Why Dirox
Our Team
Trust & Security
Partners
Dirox Gives
Careers
Locations & Contact
OFFICES
[Headquarters: city, registered address]
[Office: city, registered address]
[Office: city, registered address]
© 2026 Dirox. All rights reserved.
Legal Notice
Privacy Policy
Cookies Policy
Cookie settings
Client Support
Where does our data go, and do you train on it?
Where does our data go, and do you train on it?
Your data stays in the cloud region you choose, normally your home country. We build on Google's Vertex AI in a cloud account you own, where your data is covered by your enterprise agreement, customer-managed encryption keys, and data residency, or on open-source models in your own environment. Your data is never used to train models for anyone else, and we never sell or trade it.
Who owns the models, prompts, and code, and can we take them elsewhere?
Who owns the models, prompts, and code, and can we take them elsewhere?
You do, and you can. You own all the intellectual property we produce for you or with you: code, prompts, evaluation sets, the models we train for you, and documentation. Dirox keeps only its own tools, methods, and know-how. Third-party models stay under their providers' licenses, and we show you those terms before you commit. Everything is documented, so another team can take over.
Which models do you use, and how do we avoid lock-in?
Which models do you use, and how do we avoid lock-in?
We choose per use case, after testing on your own examples: hosted models such as Gemini on Vertex AI, or open-source models we run on your infrastructure. We put a thin layer between your application and the model, so you can switch providers without rewriting the product, and we keep the test sets that prove a new model is as good as the old one.
What will it cost to run once it's live?
What will it cost to run once it's live?
Running costs come mainly from inference (tokens or GPU time), hosting, and monitoring. We estimate them for each use case before you build, during the AI Use-Case Assessment. We keep them predictable with caching, smaller models where they are good enough, decision models, and usage limits. Once it's live, Shield Advanced or Premier reports your actual usage and cost every month.
How long does a proof of concept take, and what does it cost?
How long does a proof of concept take, and what does it cost?
Most proofs of concept run as a 2 to 4-week Studio Sprint at a fixed price, from US$X, agreed after the assessment. You see the AI working on your own data, with its accuracy measured on your own examples, before you decide on a full build. If the results don't justify one, you stop there and keep what we built.
Who works on our project, and do you use subcontractors?
Who works on our project, and do you use subcontractors?
A lead who speaks your language, English or French, and engineers in Ho Chi Minh City, Vietnam (UTC+7), all Dirox employees. We send a written update every week. If a project needs a specialist subcontractor, we name them in the proposal before work starts.
Once it's live, what support hours and response times apply?
Once it's live, what support hours and response times apply?
Every system we deliver includes Shield Essential: for 12 months from go-live, we respond to product defects within 24 hours on business days and fix them at no charge. Shield Advanced adds monitoring, security updates, version upgrades, and support with no cap on hours, and responds within 12 hours, 24/7/365. Shield Premier responds to a critical incident within 2 hours. For AI systems, Advanced also tracks token costs and checks answer quality every month.