Data and connections
A GPU server on the internal network, the model gateway, n8n and virtual machines in the client's infrastructure.
AI that keeps data inside the company
Implementation in progress. We are building an AI environment on the company's servers, with local open-weight models, so automations can work on data that cannot leave its infrastructure.
Results
- 100%*
- of production processing on the company's servers
- 2*
- first processes being prepared to run on this server
- 96 GB*
- of GPU memory for local language models
Work in progress. These are results verified at the test stage or the targets we are aiming for. We will confirm them on completion.
How it works
Data stays in the company
GPU server on the internal network
One gateway to the models
separate keys for each app
Production separate from testing
separate models and keys
n8n runs the automation
on the client’s machines
Maintenance on request only
under the client’s IT control
Company internal network
Counterparty statementGPU serverLocal model
Public cloud
no connection from outside
Gateway: app key → local model, e.g. Bielik
Statement readingCounterparty data entryNext process
Statement readingCounterparty data entry
Production
New process from a workshop
Testing
1
Document
2
Local model
via the gateway
3
Result in the system
- Maintenance request
- Client IT opens the session
- Work visible in the logs
- Session closed, no standing access
Result100%of production processing on the company's servers
01
Challenge
Data on counterparties and their financial statements may be processed in production only on the company's infrastructure, so public cloud AI services were ruled out. The company had no server for AI models, and at least two processes were waiting in line: financial statement analysis and entering counterparties into the system.
Implementation is in progress. The figures for this case study are requirements and plans we will confirm after launch.
02
Solution
We helped select a GPU server and are installing the environment on it: local open-weight models, including the Polish model Bielik, a single gateway to the models and n8n as the automation platform. We are separating the production environment from the test environment, and the models will be available only from the internal network. This server will host the financial statement solution already running in the financial analysis team and automated counterparty data entry, and we choose further processes with the teams in discovery workshops.
03
How we built it
The models will run on a GPU server in the internal network, with no external exposure, and applications will connect to them only through a gateway with separate keys and with production models separated from test models. The application layer will run on virtual machines in the client's infrastructure, and n8n will orchestrate the automations. After launch our team will have no standing access: maintenance will take place in on-request sessions under the control of the client's IT.
On-premise GPU serverOllamaBielikLiteLLMn8nPostgreSQLDocker
Workflow
How the solution works
- 01
A server inside the company
We helped choose a GPU server. The models will run on the internal network, with no external exposure.
- 02
One gateway to the models
Every application will connect to local models, including Bielik, through one gateway with its own key. Production will be separated from testing.
- 03
The first processes
The server will host the financial statement solution and automated counterparty data entry, with n8n running the automations.
The human role
The client's IT controls access. After launch we will have no standing access, service work will take place in sessions on request.
Lessons
What made the difference.
- 01
Models without the cloud
We run local open-weight models, including the Polish model Bielik, on the company's server, so data does not go to external model providers.
- 02
A single gateway to the models
Each application will connect to the models through a single gateway with its own key, and production models will be separated from test models.
- 03
Access under the client's control
After launch the contractor will have no standing access, and maintenance will be possible only in on-request sessions supervised by IT.
On-premise AILocal modelsModel gateway
Company of the Future
How this builds the Company of the Future
A project delivers results on its own, but it also leaves organized context behind in the company. Every next AI solution builds on it. This project added to the AI Engine:
- Intelligence
Local open-weight models, including the Polish Bielik, on a GPU server on the internal network.
- MCP
One gateway to the models with separate keys, and n8n for automation: the next processes connect to the same environment.
Value
Where value is created
Time saved
Each next process connects to a ready environment instead of building its own.
New possibilities
Automations can work on data that cannot leave the company.
What comes next
We choose the next processes with the teams in discovery workshops.
Most of our projects are covered by NDAs, so we only share part of them, without client names or details, as inspiration. We are happy to tell you more about our experience in a conversation.
Let’s talk about the work you want to improve.
Tell us how a process works today. Together, we will explore where AI Engine can help and which module to start with.
Book a callOr write to: hello@wearefuture.ai
