XDM as an LLM, RAG and Agentic AI Data Feeder

XDM as an LLM, RAG & Agentic AI Data Feeder

Beyond testing, XDM prepares governed datasets for analytics and AI. It provisions masked production data and production-derived synthetic datasets for customer-managed ML, LLM, and RAG pipelines. It also provisions compliant knowledge bases for enterprise agentic AI, governed, masked, and ready for agents to act on.

Clear boundary: XDM provisions the data and knowledge bases. Your teams keep the ML models, LLMs, RAG stack, vector technology, agents, and applications.

Talk to an expert Explore the XDM product

3 data approaches for AI

Masked production data, production-derived synthetic datasets, and targeted subsets, chosen per use case.

What XDM feeds

Analytics, machine learning, LLM and RAG pipelines, and the knowledge bases enterprise agents read from.

What stays yours

Model selection and training, the retrieval stack, vector technology, agent orchestration, and the applications on top.

Key Answers

Clear answers to the questions data and platform teams ask before they feed enterprise AI with production-derived data.

Why does AI need governed data provisioning?

AI initiatives stall when the only realistic data sits in production and cannot leave it safely. Governed provisioning makes that data usable outside production without exposing sensitive values.

What does XDM actually deliver?

Masked production data, production-derived synthetic datasets, and compliant knowledge bases, provisioned repeatably into the pipelines and environments your AI workloads run in.

Where does XDM stop?

At the data boundary. XDM does not train models, operate a RAG pipeline, host a vector database, or orchestrate agents. Those remain customer-managed by design.

How XDM Feeds AI Pipelines

From an AI data requirement to a governed, repeatable feed your teams can rely on.

STEP 1

Define the AI data requirement

State what the workload needs: which sources, which entities, how much realism, how much volume, and how often the feed has to refresh.

STEP 2

Choose the data approach

Masked production data, production-derived synthetic datasets, targeted subsets, or a blend, matched to the privacy and realism the use case demands.

STEP 3

Protect and provision

Sensitive values are masked before data leaves production, then delivered as a governed dataset or a compliant knowledge base for the consuming pipeline.

STEP 4

Automate the refresh

Repeat the same rule-driven workflow on every cycle, so datasets and knowledge bases stay current without ad hoc extracts.

Capabilities that make AI data provisioning governed and repeatable

The provisioning work that has to happen before any model, retrieval stack, or agent sees enterprise data.

Provision masked production data

Deliver realistic, production-derived data with sensitive values masked before it leaves its source environment.

Discuss your use case

Generate synthetic datasets

Produce production-derived synthetic data where production content should not leave its original boundary at all.

Compare approaches

Subset to what the workload needs

Keep volumes and refresh cycles practical with targeted subsets that preserve the relationships the workload depends on.

Size your feed

Build compliant knowledge bases

Provision governed, masked knowledge bases that enterprise agents can act on, prepared the same way every time.

See agentic use case

Automate provisioning and refresh

Replace manual extracts with repeatable delivery, so AI datasets and knowledge bases stay current across cycles.

Automate delivery

Govern the whole data lifecycle

Apply the same rules, masking, and controls across test data and AI data, so one governance model covers both.

Explore XDM

CHOOSE THE RIGHT APPROACH

Match the data approach to what the AI workload has to learn from.

Training, retrieval, and agent knowledge bases each need a different balance of realism, privacy, and volume. XDM supports a deliberate choice instead of handing one undifferentiated production copy to every pipeline.

Masked production data

Realistic structure and relationships preserved, with sensitive values masked before the data is used outside production.

Production-derived synthetic data

Datasets modeled on production characteristics for cases where production content must not leave its original boundary.

Targeted subsets

Smaller, relevant slices that keep feeds affordable and refresh cycles short while retaining what the workload needs.

Three ways XDM feeds enterprise AI

One provisioning platform serves three consumers. Each has a clear line between the data XDM governs and the stack you operate.

ANALYTICS & MACHINE LEARNING

ML pipelines

XDM prepares governed datasets for analytics and machine learning, provisioning masked production data and production-derived synthetic datasets on a repeatable schedule.

You manage the feature engineering, model training, and evaluation.

Discuss ML data

LLM & RAG

LLM and RAG pipelines

The same governed datasets feed customer-managed LLM and RAG pipelines, so the content entering your retrieval workflow has already been masked and governed at source.

You manage the LLMs, chunking, embeddings, vector technology, and retrieval.

Discuss RAG source data

AGENTIC AI

Agent knowledge bases

XDM provisions compliant knowledge bases for enterprise agentic AI. They are governed, masked, and ready for agents to act on.

You manage the agents, orchestration, tools, and applications.

Discuss agent data

Give your AI program data it is allowed to use.

Replace ad hoc extracts and blocked production access with governed, masked, repeatable provisioning into the ML, LLM, RAG, and agentic pipelines your teams already run.

Talk to an expert

AI DATA PROVISIONING FAQS

Questions about governed AI data provisioning

Common questions, answered.

How UBS Hainer XDM provisions governed data and compliant knowledge bases for customer-managed ML, LLM, RAG, and agentic AI work.

Ask an expert

It means XDM prepares and delivers the governed data that AI initiatives depend on. XDM provisions masked production data, production-derived synthetic datasets, and compliant knowledge bases, and hands them to the machine learning, LLM, RAG, and agentic AI pipelines your teams operate. XDM supplies and governs the data; your organization keeps ownership of the models, the retrieval stack, and the applications.

No. XDM does not train, fine-tune, host, or serve models of any kind. It provisions the datasets and knowledge bases that feed customer-managed training, fine-tuning, evaluation, and inference workflows. The choice of model, framework, and hosting remains entirely with the customer.

No. XDM does not operate a retrieval-augmented generation pipeline, and it does not host embeddings or a vector database. XDM provisions the governed, masked source content that a customer-managed RAG stack ingests, chunks, embeds, and indexes with the vector technology the customer has selected.

No. Agent orchestration, tool routing, and application logic are customer-managed. XDM provisions the compliant knowledge bases that power enterprise agentic AI. They are governed, masked, and ready for agents to act on, so the data an agent reaches for has already passed the organization’s privacy and governance controls.

XDM provisions masked production data that preserves realistic structure and relationships, production-derived synthetic datasets for cases where production content should not leave its original boundary, and targeted subsets that keep volumes and refresh cycles practical. Teams can also combine these approaches within one provisioning workflow.

Sensitive values are masked before data leaves production, and the same governed provisioning workflow is applied consistently every time a dataset or knowledge base is refreshed. Because provisioning is repeatable and rule-driven rather than ad hoc, teams can describe how each AI dataset was produced and which protections were applied to it.

It is the same provisioning platform applied to a different consumer. Test data management delivers fit-for-purpose data to development, QA, and CI/CD environments; AI data provisioning delivers governed datasets and knowledge bases to analytics, machine learning, LLM, RAG, and agentic AI pipelines. The masking, subsetting, synthetic-data, and automation capabilities are shared.

XDM is a data orchestration platform for distributed databases, including Oracle, SQL Server, PostgreSQL, and cloud targets, and it provisions into the environments where AI workloads run. For details on platform coverage for a specific program, talk to a UBS Hainer expert.