We build AI software for the public sector
And we build it the way a government agency can comfortably sign off on: data stays inside the organisation, every result traces back to a source document, and an officer always makes the final call.
Where we came from
Insightech began as a systems integrator for public sector organisations. Over six years we deployed surveillance infrastructure and IT systems for provincial departments and local authorities — work that meant operating on dedicated networks, following public investment procedures, and handing over something a local team could actually run.
That background is why we approach artificial intelligence differently from most AI companies. We did not start from "which model is the strongest", but from "how does a department in a province keep running this after we leave, without a single page of internal text going outside".
Once large language models became good enough to run on servers installed on site, we shifted our focus to building AI software for administration and information processing. Everything we now ship is designed to be installed inside the customer’s infrastructure, with no dependency on public AI services.
Company information
- Registered name
- Insightech Technology Joint Stock Company
- Tax code
- 0109320449
- Founded
- 2020
- Charter capital
- VND 10,000,000,000
- Legal representative
- Mr. Cao Tien Thong — Director
- Head office
- No. 14 Duong Thanh Street, Cua Dong Ward, Hoan Kiem District, Hanoi, Vietnam
- Team size
- 10–20 people
- Focus
- Artificial intelligence software, systems integration
Four principles that hold across every AI product we build
Our three solutions serve very different users: government agencies, citizens completing administrative procedures, and individuals managing their own knowledge. All three are built on the same four principles below. These are design constraints, not slogans — there is work we turn down because of them.
- 1
The AI runs where you control the data
Every product is designed so the AI model runs on infrastructure the user controls — your agency’s servers, your organisation’s servers, or your own machine. None of our products needs to send data to a public AI service in order to work. If a feature can only run by sending data outside, we drop the feature.
- 2
The AI only speaks within data you control
Our products do not answer from whatever the model absorbed somewhere on the internet. They answer from data loaded in under your control — an agency’s documents, administrative procedure data, or your own notes. Outside that scope the software says it has no basis to answer. We do not let it speculate to make an answer look fuller.
- 3
The AI proposes, people decide
No AI-generated result turns into a decision or an action on its own. There is always a step where the user reviews and confirms. We do not build features that let the AI act with nobody confirming, even when asked to.
- 4
No numbers promised before we measure on real data
We do not quote one accuracy rate for every case, because that number depends on the quality of the data in each place. We run a trial on your real data, measure it, and only then report the figure — and if the result is not good enough to use, we say so instead of carrying on with the sale.
How we deploy
We do not sell a package and disappear. A typical project runs through five steps, and you can stop after any of them.
- 01
Survey the current situation
A first consultation: how the organisation handles documents, assigns tasks and produces reports today; which systems are already running; what server infrastructure exists.
- 02
Trial run on real data
We take a real set of your documents and templates, run the software against them, and measure. This is where you find out whether it works for your case.
- 03
Pilot in one unit
Pick one department or one ward with a narrow scope. Low risk, and real results you can report internally.
- 04
Configure to your processes
Load your report templates, document classification rules and permission structure to match your internal regulations.
- 05
Hand over and train
Train the officers who will use it and the technical team who will run it, and hand over the system documentation. The goal is that you are self-sufficient, not dependent on us.
Why Insightech
Plenty of companies sell AI software. Here is where we differ.
We understand dedicated network environments
Deploying inside networks with no Internet access, respecting network segmentation and information security procedures — that is work we have done for years in infrastructure, not something we are learning now.
Products designed to be handed over
We build software your team can operate, with system documentation and training included, rather than locking you into the vendor.
Start small, prove it, then scale
We encourage a narrow pilot first. It lowers your risk and produces real evidence you can use to win internal support.
We tell you what we cannot do
If your problem is beyond what the technology can currently do well, we say so at the survey stage rather than promising and letting the project stall.
See it run on your own documents
Every solution sounds good in a description. The only way to know whether this one works for you is to run it against your real documents, templates and workflows. That is exactly the kind of demo we do.
The demo is free and carries no obligation. If it turns out we are not the right fit, we will say so.