Your company’s synonyms, abbreviations and approved names.
NXTGN AI · Models, open weight & operations
Decide which model processes your data.
Where AI runs, how capable the model is and what its licence permits are separate questions. Open-weight models can run locally or as a service. A frontier model can also have open weights. We test which combination of model, data paths and operations fits your tasks.
Explore scope and workflow ↓NXTGN AI CORE · Our AI product
AI using your knowledge and terminology.
A language model does not know your internal abbreviations or the relationships between your assets, components and business records. NXTGN AI CORE represents this context using a knowledge graph, glossary and taxonomy. Documents and databases supply the evidence; your specialists define terminology and relationships.
Explore NXTGN AI CORE →Relationships between terms, assets, components and sources.
Models, embeddings and OCR on your servers or hosted on a dedicated NXTGN instance.
Scope
What we deliver for you
Define data paths
Confidential content is classified before choosing a model. We document permitted processing and excluded services.
Test models
A shared set of real tasks reveals correct, incorrect and incomplete results. We also compare latency and resource requirements.
Run locally
We assess model licence, hardware, concurrent users and updates. Local deployment gives control but requires infrastructure and operations.
Connect hosted models
For external services, we assess region, contractual terms, logging and actual transmitted content. A product name alone does not answer these questions.
Control access
A shared access layer can enforce policies, budgets and model choice. Sensitive data must not fall back to another service without approval.
Check model changes
New versions are checked against the same test tasks. Changes to answers, costs and tool calls are assessed before release.
Deployment, model capability and licensing are three separate decisions.
We choose for your task. A larger model is not automatically the best solution. A locally installed model does not automatically make an application independent.
Local or as a service
Describes where processing happens and who operates it. Actual data paths across all components are what matter.
Open Weight
Model weights are available. This is not automatically unrestricted open source. Use, modification and redistribution depend on the specific licence.
Frontier
A classification of capable models, not a deployment or licence category. A model can be both frontier and open weight. Testing shows whether it handles your task better.
Background: Manufacturer example of open weight and frontier · Example of a model-specific licence
Which deployment fits your data and IT?
Your own environment
You run required components on your own or dedicated infrastructure. Independent operation also requires checking search, embeddings, document processing and telemetry for external AI dependencies.
Check first
Hardware, concurrency, model licensing, updates, backup and monitoring.
Managed on Azure
Models and applications use suitable Azure services. Regional, global and other deployment types have different processing paths; storage location is not automatically processing location.
Check first
Deployment type, data paths, identities, model availability and operating responsibility.
External model API
The application sends permitted inputs to a provider. Open-weight models can also be used through an external API.
Check first
Transmitted content, contract, retention, data use and outage behaviour.
Hybrid operation
A policy selects an approved path for the task and data class. An outage must not automatically forward sensitive content to another service.
Check first
Classification, routing, model changes, fallback and traceable logs.
Azure deployment types and processing regions ↗
The comparison starts with your test cases.
We assess business correctness, supported answers and permitted tool calls, alongside response time, concurrency, resources and behaviour when information is missing. Model, infrastructure and change rules follow those tests.
How this becomes a controlled agentic process →Model choice follows the application
We compare model capabilities, available weights and licences separately from deployment. Your task, data rules and tests determine the choice. These applications show which requirements matter.
01
Technical service knowledge
An answer sits in a manual, an old service record or a colleague’s memory. Search takes time and the document version does not always match the device.
What we deliver
We connect approved sources and organise devices, versions and terminology. AI CORE supplies answers with evidence that your team can open and check.
What you gain
Service staff find relevant information in one place. Senior colleagues spend less time answering repeated questions. Your team retains responsibility for technical approval.
How we measure value
Search time per request, answers with evidence, technically correct answers and questions escalated to experts.
02
Documents and operational workflows
Orders, inspection reports or delivery notes must be read, checked and handed over. Exceptions often end up in email or spreadsheets outside the actual process.
What we deliver
We digitise intake, data validation and approval. AI handles suitable extraction and classification. Exceptions remain visible and go to a named owner. Power Apps, the data platform and integrations implement the workflow.
What you gain
The transaction is traceable without tracking it across several lists. Your business team handles exceptions instead of repeating the same re-entry for every document.
How we measure value
Time per document, corrections, unresolved exceptions and time to business approval.
03
Customer requests and proposals
Requests arrive by email, attachments contain important details and the same data is entered again in CRM or ERP. Clarifications delay processing.
What we deliver
We review the actual workflow and connect intake, customer master data and processing. AI can classify content and prepare drafts. Rules define required fields, review and handover to your source systems.
What you gain
Your team sees open requests and works with fuller context. Less manual re-entry creates time for customer clarification and proposal quality.
How we measure value
Processing time, duplicate entry, clarification rate and time from intake to proposal.
Clarify before delivery
A local model is not automatically cheaper or safer. A frontier model is not better at every task. Your requirements and verified results determine the choice.
Do we need a GPU server?
That depends on the model, task and concurrent demand. We test resource needs before hardware is purchased.
Where does Microsoft Foundry fit?
Foundry is Microsoft's Azure platform for developing and operating AI applications with models, tools and agents. It is not a single language model. Region and capabilities are checked per service.
Can one application use both paths?
Yes, with suitable data classification, model access and tests. We also define allowed behaviour for timeouts and outages.
Let’s compare models on your tasks.
We discuss data, expected outcomes and operations to build a comparison you can assess.
