1/22 NLP architecture
2/22 NLP architecture

The NLP architecture on Google Cloud can be visualized with three layers.

3/22 NLP architecture

The bottom layer is the NLP, or in general, AI foundation, and it includes the Google Cloud infrastructure and data engineering tools.

4/22 NLP architecture

Let’s first uncover the Google Cloud infrastructure, which can be further visualized as two layers.

5/22 NLP architecture

At the base is networking and security, which is the foundation for all of Google’s infrastructure and applications.

6/22 NLP architecture

On the next layer sit compute and storage.

Google Cloud separates—or decouples, as it’s technically called—compute and storage so they can scale independently based on need.

7/22 NLP architecture

Let’s look at the compute element:

breakthroughs in this area advanced ML significantly.

In 2016, Google introduced

8/22 NLP architecture

the Tensor Processing Unit, or TPU.

9/22 NLP architecture

TPUs are Google’s custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads.

10/22 NLP architecture

TPUs act as

11/22 NLP architecture

domain-specific hardware, as opposed to

12/22 NLP architecture

general-purpose hardware

13/22 NLP architecture

with CPUs and GPGPUs.

14/22 NLP architecture

TPUs allow for higher efficiency by tailoring architecture

15/22 NLP architecture

to meet the computation needs in a domain, such as matrix multiplication in machine learning.

Cloud TPUs are integrated across Google products,

16/22 NLP architecture
17/22 NLP architecture

The other component of the NLP foundation is a set of data engineering tools.

For example, you can use Dataflow to ingest and process both batch and real-time data and use BigQuery to analyze and uncover the insights of the data.

18/22 NLP architecture
19/22 NLP architecture

Above the NLP foundation, the middle layer represents the NLP development platform, which includes three options to develop an NLP project:

  • pre-built APIs such as the the Dialogflow API and the Natural Language API,

  • AutoML, and

  • custom training with Workbench, which are offered through Vertex AI.

The top layer represents …

20/22 NLP architecture

NLP solutions, which include two groups: horizontal solutions and industry solutions.

21/22 NLP architecture

Horizontal solutions usually apply to any industry that wants to solve the same problem.

Examples include Document AI (Doc AI) and Contact Center AI (CCAI).

And the second group is vertical, or industry solutions.

22/22 NLP architecture

These represent solutions that are relevant to specific industries.

Examples include: Lending DocAI, which aims to transform the home loan experience for borrowers and lenders by automating mortgage document processing.

Retail Search, which gives retailers the ability to provide Google-quality search and recommendations on their own digital properties, thus helping to increase conversions and reduce search abandonment.