Markdown edition

# Cloudera launches Blueprints, plans NVIDIA workflows

> Cloudera launched Blueprints and described NVIDIA workflows as forthcoming.

By BIGCHANGE.AI Editorial

Published: 2026-10-29T03:11:38.761Z
Updated: 2026-10-29T03:11:38.761Z
Canonical: https://bigchange.ai/blog/cloudera-blueprints

![Conceptual illustration of an unbranded payment card beside blank receipts and a separate orange-edged receipt.](https://bigchange.ai/api/media/file/cloudera-transaction-holdout-v2.png)
AI-generated conceptual illustration by BIG CHANGE.

Cloudera launched its Blueprints portfolio on October 27. In its launch announcement, the company said NVIDIA-based workflows would be incorporated soon; it did not give a release date for each one.

## The big change

- Reference applications give data teams a starting point for building AI systems, but the work of connecting them to an organization’s own data and controls remains. A reusable pattern can make an initial evaluation more concrete without settling whether it fits the environment in which it must operate.
- NVIDIA’s separate synthetic-data example turns transaction histories into fraud-classifier features. A bank would need to compare them against its own baseline and decision requirements before drawing conclusions.
- A catalog listing describes an intended application, while a matching artifact lets a team examine the implementation and its dependencies. The transaction listing’s GitHub link led to a different project when checked, leaving the availability of that specific Cloudera implementation unconfirmed.

Cloudera’s [announcement](https://www.cloudera.com/about/news-and-blogs/press-releases/2026-10-27-cloudera-launches-cloudera-blueprints-bringing-nvidia-agent-toolkit-to-enterprise-data-anywhere.html) names planned NVIDIA workflows for finance, healthcare, telecoms and synthetic-video detection. It presents intended applications, not measured customer outcomes.

A [Cloudera catalog page](https://www.cloudera.com/solutions/ai-blueprints/cloudera-nvidia-vast-building-foundation-models.html) describes a transaction model using Cloudera AI Workbench, VAST storage and NVIDIA integration. It says Blueprints ship with code, sample data and instructions.

When BIG CHANGE checked the page on October 29, its GitHub link led instead to a [network lakehouse repository](https://github.com/cloudera-applied-ai/AI_Blueprint_Network-Lakehouse-for-Agentic-Operations), whose stated purpose is network operations. 

## From transaction history to a fraud feature

In a separate [NVIDIA developer example](https://developer.nvidia.com/blog/build-your-own-transaction-foundation-model-for-financial-intelligence/), transaction records become token sequences used to train a next-token prediction model. The resulting representation, or embedding, condenses information from a transaction history into numbers that another model can use.

NVIDIA reports that an XGBoost fraud classifier combining raw features and embeddings reached average precision of 0.1755, versus 0.1238 for raw features alone, on a 100,000-row test sample. That is a 41.76% relative increase, not percentage points. The synthetic IBM TabFormer data were split chronologically, and the test sample retained roughly 0.1% fraud prevalence. Average precision summarizes precision across recall levels; it does not establish an improvement on a bank’s own data or a Cloudera deployment.

NVIDIA publishes a [separate transaction-model repository](https://github.com/NVIDIA-AI-Blueprints/transaction-foundation-model). BIG CHANGE reviewed the public records but did not install or test either implementation.

## What an evaluation would need to establish

A [Dell technical paper](https://www.dell.com/en-us/infohub/articles/transaction-foundation-models-on-dell-ai-factory-a-new-intelligence-layer-for-financial-services) reports its own run of the NVIDIA example and recommends a time-ordered holdout. Keeping later transactions for evaluation gives the comparison a clearer temporal boundary than mixing future records into training. The paper also recommends measuring the full inference path, including retrieval of transaction history: timing the model alone would miss work needed to produce the features.

For a financial institution, the relevant comparison is whether the additional representation improves decisions against its existing baseline under its own data and business thresholds. It also needs to establish who can access the histories and how the required components fit its infrastructure. Those are evaluation questions, not results reported by the launch.

## Sources

- [Cloudera Blueprints launch](https://www.cloudera.com/about/news-and-blogs/press-releases/2026-10-27-cloudera-launches-cloudera-blueprints-bringing-nvidia-agent-toolkit-to-enterprise-data-anywhere.html) — Launch announcement and planned workflows.
- [Cloudera transaction blueprint listing](https://www.cloudera.com/solutions/ai-blueprints/cloudera-nvidia-vast-building-foundation-models.html) — Transaction blueprint listing and GitHub link.
- [Cloudera network lakehouse repository](https://github.com/cloudera-applied-ai/AI_Blueprint_Network-Lakehouse-for-Agentic-Operations) — The linked repository describes a network operations lakehouse.
- [NVIDIA transaction-model example](https://developer.nvidia.com/blog/build-your-own-transaction-foundation-model-for-financial-intelligence/) — Transaction modeling with synthetic data.
- [NVIDIA transaction-model repository](https://github.com/NVIDIA-AI-Blueprints/transaction-foundation-model) — Separate NVIDIA reference implementation.
- [Dell transaction-model technical paper](https://www.dell.com/en-us/infohub/articles/transaction-foundation-models-on-dell-ai-factory-a-new-intelligence-layer-for-financial-services) — Dell describes its example run and evaluation considerations.