On-premises deployment · Industry systems · Ongoing operations

Run AI inside your own network

Models run on servers the customer designates. The applications cover medical education, research assets, knowledge Q&A, and field inspection. Where data lives, who can access it, and how acceptance is measured are confirmed before the project starts.

Start a conversation See the main work

On-premises deployment On the private network

User side Teacher / learner browser

HTTPS

Entry tier · Server A

Nginx reverse proxy Vue 3 front end

Private network /api/

Core tier · Server B (GPU)

Private vLLM model FastAPI backend PostgreSQL MinIO object storage

The entry point, models, business systems, and data can be placed on the customer's own network.

Scope of work

What we deliver is a system that keeps running, not a one-time model install.

Servers, models, data, permissions, interfaces, and operations are designed together. Where a small trial is practical, we start there, then move into full delivery and accept the work against the agreed list and measures.

01 · Private AI

On-premises AI deployment and system integration

Models, knowledge bases, and applications are deployed in the environment the customer specifies, matched to their data-management rules and the systems they already run. Permissions, logs, monitoring, and external interfaces are set up with the deployment. Materials needed for security review and day-to-day operations are included.

Capacity and storage planning Models and knowledge bases in place Accounts, permissions, and runtime logs Connections to existing business systems

Compute GPU servers and storage

Data Databases, object storage, and knowledge bases

Model Large language models, RAG, and vision models

Application Q&A, workflow assistants, and business systems

02 · Medical education

Simulated consultation and clinical-skills training

Simulated cases cover consultation, history taking, clinical reasoning, and communication with patients. Learners can repeat a case. Each conversation is kept and scored against one standard, and teachers review the session with the class. The system is for teaching. It is not used for real diagnosis or treatment.

Simulated cases and branching dialogue A full record of each consultation Scoring against one standard Teacher review and class-level patterns

Training flow Every exercise leaves a record that can be reviewed

01
Case setupTeachers set the role, symptoms, key questions, and scoring rules
02
Learner consultationQuestions and follow-ups follow a real pace and build a diagnostic line of thought
03
In-session scoringCompleteness, clinical reasoning, and communication are recorded
04
Teacher reviewKey exchanges, weak points, and patterns across the class

Consultation completeness Red-flag symptoms Clinical reasoning Clinician–patient communication

03 · University research

Zhiyan Yunshu: research asset management

For university research groups, documents, data, and code that sit with different people are brought into one place, with versions, owners, and handoffs recorded. Supervisors can see progress, and there is a record when a member graduates or leaves the group.

Documents, data, and code in one place Versions and owners can be traced Transfer, archive, and departure handoff AI summaries, questions, and risk notes

04 · Field inspection

Intelligent inspection in the field

For transmission towers, roads and bridges, campuses, and plant facilities, drones, on-site capture devices, and image recognition are connected to the inspection process already in use. After capture, suspected defects are marked automatically. A person then reviews them, assigns a severity, and issues the report.

Drone and on-site capture Automatic defect marking Local storage when the site has no network Human review, grading, and the report

Sample inspection of corrosion on an engineering structure

From a site photo to a defect record that can be reviewed Prototypes and pilots are available for transmission towers, roads and bridges, and plant facilities.

How a project moves

Scope and acceptanceare set before work starts.

A first engagement can be a small trial: one department, one course, one class of documents, or one set of site photos. After the result, the data conditions, and the way of working together are confirmed, full delivery follows the customer's budget and procurement rules.

For projects that use public funds, special-purpose funds, or an annual budget, we can help prepare the requirements, data conditions, delivery list, and acceptance measures. The customer decides the procurement method under their organization's current rules.

01Confirm the needScenario, department, and data conditions

02TrialA limited scope, to see the result first

03Proposal and budgetDeliverables, resources, and schedule

04Approval and procurementUnder the customer's own rules

05Delivery and acceptanceTesting, training, and handover of materials

06Annual operationsMonitoring, upgrades, and issue handling

Technical services that can be contracted on their own

Model and RAG evaluationTest sets, performance tests, and acceptance records

Domain knowledge basesOrganizing source material, retrieval, and source tracing

On-premises deployment and integrationModels, permissions, interfaces, and the runtime environment

Annual system operationsMonitoring, incident handling, upgrades, and service reports

How we work

Proposal, deployment, testing, and later operations stay with the same team.

The same point of contact stays for the whole project. Deployment, testing, training, and issue handling during implementation are recorded, so they can be looked up and maintained later.

01 A clear deployment boundary

Where the model runs, where data is stored, who can access it, and which interfaces are open are set from the customer's existing rooms, servers, and network.

02 Existing systems stay in place

Systems already in use keep running. Knowledge Q&A, training assessment, or image recognition is added through interfaces.

03 The handover package is complete

Deployment records, test materials, operating notes, and a rollback plan are handed over together. Upgrades and annual operations can continue after that.

Partners

Work with regional partners to supply the AI portion of a project.

For regional software companies, systems integrators, and hardware vendors, we share scenario analysis, AI module integration, on-premises deployment, implementation, and after-sales support. Who owns the customer relationship, how the commercial work is split, and how responsibility is divided are confirmed before the cooperation starts.

Discuss a partnership

01Joint proposal

Agree the requirements, the technical approach, and the split of work.

02AI modules

Knowledge bases, model services, training assessment, and image recognition.

03Implementation support

Deployment, interface work, testing, training, and project acceptance.

04After-sales operations

Issue handling, version upgrades, and longer-term service, as the contract states.

Contact

Projects in on-premises AI, medical education, research assets, or field inspection can start with one specific scenario.

Tell us the department, the systems already in use, the data conditions, and the planned timing. We can then judge which scenario to start with.

Phone 187-6588-0081 Email liguoma@horsduroot.com