Vertex AI Agent Builder - GDG Alicante - Julio 2024

Evento sobre Vertex AI Agent Builder del GDG Alicante

1. Vertex AI Agent
Builder

2. Table of Contents


01
02


GDG Alicante → Intro
Vertex AI Agent Builder

3. Section 1
Comunidad abierta y colaborativa
GDG Alicante

4. ¿Quienes somos?
Nicolás López
Cloud Architect
Devoteam G Cloud
Marcos Remon
Data Tech Lead
Devoteam G Cloud
Carlos Conca
Data Engineer
Devoteam G Cloud
Jesús Llor
Technical Lead
EUIPO

5. Section 1
●Sin ánimo de lucro.
●Objetivo de mejorar la comunidad.
●Entusiastas tecnológicos.
●Aumentar el conocimiento.
●Crear relaciones.
¿Qué es es un GDG?

6. Eventos, conferencias, talleres, libres y gratuitos en Alicante
Que queremos hacer
Section 1

7. Section 1
●Gemini Pro con 2 millones de tokens (2h video, 22 h audio, 60.000+
líneas de código, 1,4 millones de palabras).
●Gemma v2 con 9B y 27B de parámetros → mejora sustancial de
rendimiento.
●Gemini API disponible → context caching, code execution.
(goo.gle/cookbook ai.google.dev)
●Google Checks → IA para análisis compliance en código
(checks.google.com)
●AI Studio → aistudio.google.com
●AI Edge → Framework para desarrollo de aplicaciones con modelos
locales.
●Chrome DevTools Console insights → Asistencia de Gemini a DevTools
Console (proximamente)
Google IO Connect (Berlin) → Novedades más interesantes

8. Section 2
(f.k.a. Vertex AI Search & Conversation)
Vertex AI Agent Builder

9. Vertex AI Agent Builder value proposition
SOTA performance
Vector Search with
Vertex AI
Embeddings to
build custom
Search, Recs or
other gen AI
application

Simple UI so
developers with
minimal coding
Years of Google’s
research and
experience in
search

Understanding of
relationships with
Google’s KG

LLM to inform
relevance, ranking
and personalization
Platform for
production-ready
RAG applications
powering variety of
use cases (all
RAG/Grounding -
search)

Simplified process
of building search
applications in a
few clicks
A production-ready
Grounding system
Best-in-class
search quality
Control for
advanced
applications
Enterprise Readiness
(regional availability,
languages support)
Compliance and
Security standards
(e.g., HIPAA, ISO
27000-series, and
SOC-1/2/3, VPC-SC
and CMEK

Differentiated
grounding
capabilities
Grounding on your
own private data
with summaries and
citations alongside
the results

Grounding on
Google.com Search
Differentiated
grounding
capabilities
2 31
4 5

10. Ground & AugmentOrchestrate
Create, launch, and manage
your agents at scale

Increase generative AI output
accuracy and relevancy
Use Tools
Ground with Google
Search to access fresh, high
quality information
Public Preview

Ground on your own enterprise
data quickly with out-of-the-box
RAG in Vertex AI Search
Generally Available

Build DIY RAG providers with
LlamaIndex on Vertex
Public Preview
Connect LLMs to external tools;
call APIs and Services
Build at any level: no code
1
, low
code, or full code options in
Vertex AI Agent Builder &
Agent API

Create your own actions with
Function Calling
accessing custom or private APIs
Generally Available
Deploy and orchestrate
custom agents with
LangChain on Vertex
Public Preview


Access pre-built reusable modules
with Extensions
2

Public Preview


Vertex AI Agent Builder Enterprise Ready Tools
Develop and deploy agents faster, grounded in your enterprise truth

1
Using natural language (Public Preview)
2
Two available: Code Interpreter and Vertex AI Search

11. 011ProprietaryGoogle Cloud Next ‘24
Configure production ready agents
via a convenient SDK, add tools, add
instructions and examples, evaluate
and test your agent. The Agent API
provides a fully managed Agent
building and hosting service; your
interface is colab, python code, or
any application you want to build.
No Code Code-first
Build Your Own ML/AI to leverage your
proprietary institutional knowledge

Model training/ tuning
Build agents from the ground up, with
whatever software you like, and orchestrate
any components you develop or link in
Vertex AI APIs as tools. Create and run your
agent on Langchain on Vertex (SDK +
runtime) or LlamaIndex on Vertex (API) or
build and deploy agents with Google’s 1st
party framework for app builders Genkit
(TS/JS).
Non Developers
No ML skills required
Developers
Advanced ML skills
Agent builder products for different needs
Agent Builder in the console allows you
build high quality, out-of-box agents with
natural language describing process steps
and examples for use in conversational and
other interfaces.

Fast, easy - build, test, deploy in
minutes/hours.
Low Code
SOTA observability, experimentation and analysis tools to debug, evaluate and optimize agent & model outputs and optimize for quality

12. Best models from Google & the
industry
End-to-end model building
platform with choice at every level
Develop and deploy agents faster,
grounded in your enterprise truth
Built on a foundation of enterprise
readiness
Gemini 1.0 Pro
GA
and Gemini 1.5
Pro
Public Preview
Imagen 2.0 Editing | Live Images |
Digital Watermarking
Gemma: CodeGemma and
RecurrentGemma
Hugging Face integration
Prompt Management
Gen AI Evaluation
Grounding on Google Search
Public Preview

Vertex AI Agent Builder
Retrieval Augmented
Generation (RAG): DIY
Digital Watermarking
Data residency and expanded
regional availability + ML
processing
Vertex Embeddings Model
Claude 3: Haiku | Sonnet
Claude 3: Opus (coming soon)
Chirp 2

Supervised Tuning for Gemini Pro
Feature Store 2.0
Pipelines Integration w/ Dataplex
Lineage
LangChain on Vertex AI
Extensions
Function Calling
Search Primitives
Connectors
Responsible AI (RAI) Tooling
Org policy for model access (IAM)
Vertex AI Agent Builder | Enterprise Ready Generative AI

13. The Agent Opportunity

14. Generative AI is
transforming how we
interact with technology

15. Proprietary & Confidential
015
Agents present the biggest opportunity to drive business
impact with AI
101 real world gen ai use cases

16. Potential for transformation across many functions and industries


Customer agents:
●[Support] Product issue troubleshooting, FAQ
●[B2C/eCommerce] Product catalog, cart, orders
●[B2B] Configure, place, manage orders
●[Travel] Discover experiences, travel booking
Employee agents:
●[HR] Employee onboarding, Benefits enrollment
●[Sales] Opportunity logging
●[Payables] Invoice Payments
●[Supply chain] Inventory, Fulfillment, tracking
Knowledge agents:
●[Specialized] Legal, R&D, Financial Services, Mkt Research
●Knowledge workers
●Customer Support & Sales
Voice agents:
●Contact center
●PoS / Food ordering - drive-throughs, retail
●Automobiles - navigation, support, diagnostics
Employee agents:
●[HR] Employee onboarding, Benefits enrollment
●[Sales] Opportunity logging
●[Payables] Invoice Payments
●[Supply chain] Inventory, Fulfillment, tracking
Knowledge agents:
●[Specialized] Legal, R&D, Financial Services, Mkt Research
●Knowledge workers
●Customer Support & Sales
Voice agents:
●Contact center
●PoS / Food ordering - drive-throughs, retail
●Automobiles - navigation, support, diagnostics
Customer agents:
●[Support] Product issue troubleshooting, FAQ
●[B2C/eCommerce] Product catalog, cart, orders
●[B2B] Configure, place, manage orders
●[Travel] Discover experiences, travel booking

Customer supportDocument Search Market researchI need to
assess the
semicondu
ctor
industry
Car rental
Retail
Healthcare claims
Flights
Telecom
Payment arrangement

17. What is an “AI Agent”?
An application that reasons on how to best achieve a goal based on
inputs and tools at its disposal
Generative AI Models
(An agent can use
multiple models)
APIs Functions
Databases Agents
Tools
Profile, goals &
instructions
short-termlong-term
Memory
Model based
Reasoning/Planning
(Question Decomp &
Reflection)
Agent
Orchestration
(e.g. Agent Brain)
Key Components
●Model: Used to reason over goals,
determine the plan and generate a
response
●Tools: Fetch data, perform actions
or transactions by calling other
APIs or services
●Orchestration: Maintain memory
and state (including the approach
used to plan), tools, data
provided/fetched, etc

18. There is a spectrum of agent types for different use cases
Deterministic Agents Generative AgentsHybrid Agents
Workflow based agent that has predefined
paths and actions. Typically uses event
driven behavior. and offers higher control
and predictability. Example: basic
customer support agent.
Benefits
●Higher control and predictability of
outcomes
●More mature technology
Challenges
●Every workflow path needs to be
defined and maintained
●Lack of reasoning to creatively
problem solve outside predefined
configuration

Combines workflow based and generative
capabilities. For example, generative
capabilities can be used where there is
not a predefined workflow action.
Benefits
●Leverage predictability of
deterministic components
●Use probabilistic capabilities when
pre-defined rules are exhausted
Challenges
●Increased complexity maintaining
broader technology stack


AI powered agent driven by defined goal
and the observed environment. Reason
through future options to assess the
impact of these actions on their goals.
Benefits
●Can tackle complex problems where
clear cut rules do not easily apply
●Increased adaptability and flexibility
Challenges
●Evolving technology
●Require guardrails to use safely
and responsibly

19. Agent architectures vary broadly in complexity and architecture
Single agent Architecture Multi-Agent Architecture
Powered by a single LLM that performs all the
reasoning, planning and actions. Simplest architecture
to set up. Agent is given instructions, and tools to
reach given goal.
When to use: When process is well defined and list of
tools is limited in scope
Benefits
●Easier to implement
Challenges
●More prone to get stuck in execution loop




Powered by two or more agents that can be used for to
coordinate, collaborate & specialize which can be
organized Horizontally or hierarchical
When to use: When feedback from multiple agents is
beneficial or when parallelization is desired
Benefits
●Use specialized agents for specific tasks and to
drive efficiencies
Challenges
●More complex to setup and maintain
●Horizontal architectures can lead to group chat and
loss of focus
●Vertical architectures susceptible to leading agent
not sending critical information to other agents
Hierarchical
Horizontal

20. Why Agents on the Vertex AI Platform

21. Vertex AI is built for developers
Extensive quick start library with code samples
and jumpstarts for developers of all levels and
ecosystems
Developer labs and training resources across
Vertex products at Cloud Skills Boost at no
additional cost
Robust integrations with popular third-party
developer tools like Lang Chain, LlamaIndex,
Pinecone, and Weaviate.
Packages and extensions to natively support
Google Cloud foundation models in Google app
developer frameworks like Firebase and Flutter.
Colab Vertex AI
Firebase Flutter
Interfaces for
all developers

22. It takes more than just a model to drive business value with GenAI
Google Cloud Infrastructure (GPU/TPU) | Google Data Cloud
Vertex AI Model Garden

Google | Open | Partner
Vertex AI Agent Builder

OOTB and custom Agents | Search
Orchestration | Extensions | Connectors | Document Processors | Retrieval engines | Rankers | Grounding
Vertex AI Model Builder

Prompt | Serve | Tune | Distill | Eval | Notebooks l Training | Feature Store | Pipelines | Monitoring
AI Solution

Contact Center AI | Risk AI | Healthcare Data Engine | Search for Retail, Media and Healthcare
Build your own generative AI-powered agent
Gemini for Google
Cloud
Gemini for Google
Workspace

23. Google Cloud Infrastructure (GPU/TPU) | Google Data Cloud
Vertex AI Model Garden

Google | Open | Partner
Vertex AI Agent Builder

OOTB and custom Agents | Search
Orchestration | Extensions | Connectors | Document Processors | Retrieval engines | Rankers | Grounding
Vertex AI Model Builder

Prompt | Serve | Tune | Distill | Eval | Notebooks l Training | Feature Store | Pipelines | Monitoring
AI Solution

Contact Center AI | Risk AI | Healthcare Data Engine | Search for Retail, Media and Healthcare
Build your own generative AI-powered agent
Gemini for Google
Cloud
Gemini for Google
Workspace
Vertex AI

24. Google Cloud Infrastructure (GPU/TPU) | Google Data Cloud
Vertex AI Model Garden

Google | Open | Partner
Vertex AI Agent Builder

OOTB and custom Agents | Search
Orchestration | Extensions | Connectors | Document Processors | Retrieval engines | Rankers | Grounding
Vertex AI Model Builder

Prompt | Serve | Tune | Distill | Eval | Notebooks l Training | Feature Store | Pipelines | Monitoring
AI Solution

Contact Center AI | Risk AI | Healthcare Data Engine | Search for Retail, Media and Healthcare
Build your own generative AI-powered agent
Gemini for Google
Cloud
Gemini for Google
Workspace
Develop and deploy
agents faster,
grounded in your
enterprise truth

25. 025Google Cloud Next ‘24 Proprietary
No code Low code High code
Vertex AI Agent Builder
Tools
Orchestration
Extensions
Function Calling
Connectors
Search Primitives
Private datasets
Public datasets

Langchain
GenKit
Doc processors
Retrieval engines
Rankers
….
Freshness & Factuality Augmentation & Action
Google Cloud Infrastructure/Google Data Cloud
Vertex AI Model Garden
Vertex AI Model Builder
LlamaIndex
No Code Low Code Full Code
ALTERNATIVE OPTION








Agents

DIY


and Augment




Connectors
Take Action



Extensions

Function Calling
Primitives



Doc processors

Vector Search

Rankers and more
No code Low code Full code

26. THE VALUE OF THE
VERTEX AI TOOLSET

27. Ground & AugmentOrchestrate
Create, launch, and manage
your agents at scale

Increase generative AI output
accuracy and relevancy
Take Action
Ground with Google
Search to access fresh, high
quality information
Public Preview

Ground on your own enterprise
data quickly with out-of-the-box
RAG in Vertex AI Search
Generally Available

Build DIY RAG providers with
LlamaIndex on Vertex
Public Preview
Connect LLMs to external tools;
call APIs and Services
Build at any level: no code
1
, low
code, or full code options in
Vertex AI Agent Builder

Create your own actions with
Function Calling
accessing custom or private APIs
Generally Available
Deploy and orchestrate
custom agents with
LangChain on Vertex
Public Preview


Access pre-built reusable modules
with Extensions
2

Public Preview


Vertex AI Agent Builder Enterprise Ready Tools
Develop and deploy agents faster, grounded in your enterprise truth

1
Using natural language (Public Preview)
2
Two available: Code Interpreter and Vertex AI Search

28. get_route = Extension.create(manifest = {
"name": "get_route_tool",
"description": "Extension to get running route for a given neighborhood",
"api_spec": { [openapi specs] },
"auth_config": {"auth_type": "GOOGLE_SERVICE_ACCOUNT_AUTH"},
},
)
cymbal_running = Agent(
display_name = "Cymbal Running Agent" ,
description = "You help Cymbal Running customers find the best routes and gear." ,
instructions="""
Step 1. Confirm the location and mileage of the run
Step 2. Ask the runner where they’d like start their run
…
Step 6. If the runner asks for gear recommendations based on run, route runner
to ${AGENT:gear_recos}...""" ,
tools = [get_inventory, get_order_status],
model = ”gemini-pro”
)
Agent Builder Agent API: a flexible low code
interface to build agents

Low CodeOrchestration
Vertex AI Agent API / Agent SDK

Fully managed Agent runtime.
Use an API to add tools and instructions to Agent.
Expose Agent via integrations or API calls.

Generative AI
Models
(An agent can use
multiple models)
APIs Functions
Databases Agents
Tools
Profile, goals &
instructions
short-term long-term
Memor
y
Model based
Reasoning/Planning
(Question Decomp &
Reflection)
Agent
Orchestration
(e.g. Agent Brain)

29. Agent Builder Platform: assemble an agent from
foundational components on any
framework/language.
Agent Builder
(abstract)
Code first
Genkit
Firebase Gen AI TS/JS framework
An open source AI integration
framework that empowers app
developers to build features
powered by Google’s GenAI
models, Google Cloud
services, and their business
data.
go/what-is-genkit
Orchestration
LangChain
with/out LangChain on Vertex
LangChain is a Python
framework designed to
streamline AI application
development, focusing on
real-time data processing and
integration with Large Language
Models (LLMs). It offers an
agents module among many
other component
Anything else…
platform supporting any user
Developers know how to call
APIs, and assemble applications.
Many customers explore in one
framework and rebuild for
production in another.

Ultimately our platform will
support any customer who can
make an API call, and we have
simplified integrations and well
lit paths for popular languages
and frameworks.

30. Google Cloud Next ‘24 Proprietary
LangChain on Vertex AI
1.Deploy your OSS LangChain on Vertex AI
When you need orchestration for agent-like behavior, use
can implement them with LangChain

2.Support a wide ranges of agent tasks,
including RAG
From simple tasks like information retrieval and RAG, to
complex workflow for customer support

3.Full control in development
Define functions, tools, and workflows, and let Gemini
handle the selection of appropriate API calls and extracting
parameters from prompts.

def get_exchange_rate(
from: str = "USD",
to: str = "EUR"):

"Retrieves the exchange rate."

response = requests.get(
"https://api.currency.app/",
params={"from": from,
"to": to})

return response.json()
app = llm_extension.LangchainAgent(

tools=[get_exchange_rate],

model_kwargs={
"temperature": 0.3,
"top_p": 1,
safety_settings: {...},
}
)
remote_app =
reasoning_engines.ReasoningEngine.create(
LangChainAgent(),
requirements=[
"google-cloud-aiplatform",
"langchain",
"requests==2.*"])

remote_app.query(
query="What's the exchange rate
from US dollars to Swedish
currency?")

Code firstOrchestration
Define
function(s)
Use LangChain
templates
Deploy to
Vertex AI
Now available in Public Preview

31. Vertex AI Agents provides a no-code interface
to build production ready agents
Agent Builder
@ Generally Available
Orchestration No Code
Description
Instructions
Tools
High-level goal
Step-by-step instructions on how
the agent should behave & reason
Set of tools the agent can leverage

32. Ground & AugmentOrchestrate
Create, launch, and manage
your agents at scale

Increase generative AI output
accuracy and relevancy
Take Action
Ground with Google
Search to access fresh, high
quality information
Public Preview

Ground on your own enterprise
data quickly with out-of-the-box
RAG in Vertex AI Search
Generally Available

Build DIY RAG providers with
LlamaIndex on Vertex
Public Preview
Connect LLMs to external tools;
call APIs and Services
Build at any level: no code
1
, low
code, or full code options in
Vertex AI Agent Builder

Create your own actions with
Function Calling
accessing custom or private APIs
Generally Available
Deploy and orchestrate
custom agents with
LangChain on Vertex
Public Preview


Access pre-built reusable modules
with Extensions
2

Public Preview


Vertex AI Agent Builder Enterprise Ready Tools
Develop and deploy agents faster, grounded in your enterprise truth

1
Using natural language (Public Preview)
2
Two available: Code Interpreter and Vertex AI Search

33. Your data
Use Grounded
Generation API to
ground in your own data.
Google Search
Ground your response in
the world's knowledge
with Google Search
grounding for Gemini.

Vertex AI provides comprehensive Grounding
solutions
RAG Validation
& Analysis
Use Check Grounding to
validate grounding of a
generated response.

World knowledge
RAG
Orchestration
Use the LlamaIndex and
Langchain to build
custom RAG workflows

Private knowledge
Grounding

34. 1.Anchor model responses
LLM response directly based on trusted
Google Search world knowledge and
public facts
2.Source data provided, reducing
hallucinations
Grounding provides citations and
attributions, so you know where the
information is coming from, and can more
easily identify hallucinations
3.Works out-of-box without any
development needed

Enable the feature with simple
configuration to different data sources

Grounding on Google Search

Ground model responses in Google Search providing access to fresh, high-quality information that significantly improves accuracy of responses
Google Search Grounding provides
source links for LLM answers as well as
suggested searches so users can
quickly verify the LLM response and
continue their information journey

●Search Grounding: Tap inline
source links to open the source
webpage

●Suggested Searches: Tap Google
Search suggestion to open the
search results page






Public Preview
Allow List
1
1
Requires Opt-In
Ground with Google Search
@ Public Preview
No codePublic DataGrounding

35. ServingValidateRank
Index /
Retrieve
Embed
Collection
(web, files, DBs,
connectors, etc.)
Process &
Annotate
Generate
Parsing, chunking, embedding, indexing/ storage, semantic+token-based ranking, query understanding, user events …

Collect Build
Vertex AI Search: OOTB
Collection
(web, files, DBs,
connectors, etc.)
Serving
Run/ Serve
Vertex AI Search
Prebuilt RAG Search to simplify application build
Ground with Google Search
@ Public Preview
No codePrivate DataGrounding

36. Grounding on Vertex AI: Bringing enterprise truth to your GenAI


Grounding with
Google Search
GENERALLY
AVAILABLE
Grounding with
3P data
Coming Next
Quarter
Only provider to offer grounding
with Google Search

Available with Gemini models
only
Currently working with premier
providers such as


Grounding with
high-fidelity
Experimental
With fine-tuned Gemini 1.5
Flash model

Uses only provided context to
generate answers

Ensures high levels of factuality
in response
New New

37. Grounding with High Fidelity: Introducing grounding scores and
sourcing from provided context


Given context/Input:
Alphabet quarterly and annual reports
Google's revenue in Q1 2024 was $80.5 billion, which
represents a 15% year-over-year growth.

Grounding Score: 99.2%,
Source: 2024q1-alphabet-earnings-release-pdf
(Page 1)
Prompt: What was Google’s Q1 2024 revenue?
What was YoY growth?
The provided sources only contain financial
information for Alphabet Inc. for Q1 2024 and
previous quarters, but do not include any information
about Google's revenue for Q3 2024.

Grounding Score: 3%
Prompt: What was Google’s Q3 2024 revenue?

38. 1.Anchor model responses
LLM response directly based on your own
data (private or customer controlled) with
your own DIY RAG Semantic Search built
with LlamaIndex hosted on Vertex
2.Source data provided, reducing
hallucinations
Grounding provides citations and
attributions, so you know where the
information is coming from, and can more
easily identify hallucinations
3.Compatible with LlamaIndex

Build your own RAG with LlamaIndex and
make it enterprise ready on Vertex

Grounding on LlamaIndex on Vertex

Ground any model responses in LlamaIndex on Vertex providing access to fresh, high-quality private information that significantly improves
accuracy of responses
LlamaIndex on Vertex Grounding
provides source links for LLM answers
as well as the full search results so
users can quickly verify the LLM
response and developers can do more
with results

●Grounding Citations to your data:
Tap inline source links to open the
source webpage, document, etc
(you can control the URLs)

●Search Results: All search results
are returned along with summaries,
so you can use the metadata or
chunks for your own purposes
Code firstPrivate DataGrounding

39. Google Cloud Next ‘24 Proprietary
LlamaIndex on Vertex AI
1.DIY - Built to be fully customizable
Use default settings to start with, or configure your own
for every step, from chuck size, chuck overlap, choice of
embedding models, vector database, and many more.
2.Enterprises-grade RAG at scale
Enterprise-ready features to support large number of
documents, and wide range of connectors of data
sources
3.Easy, work out-of-box
Managed API enables developers to build RAG
applications with as little as 4 lines of code











# Create RagCorpus
rag_corpus =
rag.create_corpus(display_name=display_name)

# Import Files to the RagCorpus
response = rag.import_files(
rag_corpus.name,
paths,
chunk_size=512,
chunk_overlap=100,
# Create a RAG retrieval tool
rag_retrieval_tool = Tool.from_retrieval(
retrieval=rag.Retrieval(
source=rag.VertexRagStore(
rag_corpora=[rag_corpus.name],
similarity_top_k=3,
),
)
)

# Generate response
response = rag_model.generate_content("What is
RAG and why it is helpful?")
print(response.text)



Now available in Public Preview
Code firstPrivate DataGrounding

40. Why check Grounding?
– Grounding of the response
– Attribution to sources


Facts Answer Gen Answer
Retrieve &
Rank ?
RAG
Private DataGrounding Low Code

41. Check Grounding example
Private DataGrounding Low Code
Facts &
Instructions
Answer gen Answer
Check
Grounding
New!
Confidence
scores &
Attribution to
sources
LLM prompt for the user question:
Tell me about the movie Inception.

Retrieved facts:
●[1] Inception is a 2010 science fiction action film
written and directed by Christopher Nolan, who
also produced the film with Emma Thomas, his
wife. […]
●[2] Inception stars Leonardo DiCaprio, Ken
Watanabe, Marion Cotillard, Joseph Gordon-Levitt,
Elliot Page, Tom Hardy, Cillian Murphy, Tom
Berenger, Dileep Rao, and Michael Caine among
others. […]
●[3] Inception's premiere was held in London on July
8, 2010; it was released in both conventional and
IMAX theaters beginning on July 16, 2010.
Inception grossed over $837 million worldwide,
becoming the fourth-highest-grossing film of
2010. […]
●[4] The movie Inception is about a professional
thief who steals information by infiltrating the
subconscious of his targets.

42. Check Grounding example
Private DataGrounding Low Code
Facts &
Instructions
Answer gen Answer
Check
Grounding
New!
Confidence
scores &
Attribution to
sources
LLM prompt for the user question:
Tell me about the movie Inception.

Retrieved facts:
●[1] Inception is a 2010 science fiction action film
written and directed by Christopher Nolan, who
also produced the film with Emma Thomas, his
wife. […]
●[2] Inception stars Leonardo DiCaprio, Ken
Watanabe, Marion Cotillard, Joseph Gordon-Levitt,
Elliot Page, Tom Hardy, Cillian Murphy, Tom
Berenger, Dileep Rao, and Michael Caine among
others. […]
●[3] Inception's premiere was held in London on July
8, 2010; it was released in both conventional and
IMAX theaters beginning on July 16, 2010.
Inception grossed over $837 million worldwide,
becoming the fourth-highest-grossing film of
2010. […]
●[4] The movie Inception is about a professional
thief who steals information by infiltrating the
subconscious of his targets.

Inception was directed by
Christopher Nolan and he was also
one of the producers.

According to Wikipedia, it is about a
thief who steals information by
getting into people's dreams.

It received five Academy Awards.

It starred Leonardo DiCaprio and
Marion Cotillard among others.

It made more than $950 million in
revenue at the box office.
Correct statements in green, manually added
wrong statements in red.

43. Check Grounding example
Private DataGrounding Low Code
Facts &
Instructions
Answer gen Answer
Check
Grounding
New!
Confidence
scores &
Attribution to
sources
LLM prompt for the user question:
Tell me about the movie Inception.

Retrieved facts:
●[1] Inception is a 2010 science fiction action film
written and directed by Christopher Nolan, who
also produced the film with Emma Thomas, his
wife. […]
●[2] Inception stars Leonardo DiCaprio, Ken
Watanabe, Marion Cotillard, Joseph Gordon-Levitt,
Elliot Page, Tom Hardy, Cillian Murphy, Tom
Berenger, Dileep Rao, and Michael Caine among
others. […]
●[3] Inception's premiere was held in London on July
8, 2010; it was released in both conventional and
IMAX theaters beginning on July 16, 2010.
Inception grossed over $837 million worldwide,
becoming the fourth-highest-grossing film of
2010. […]
●[4] The movie Inception is about a professional
thief who steals information by infiltrating the
subconscious of his targets.

Inception was directed by
Christopher Nolan and he was also
one of the producers.

According to Wikipedia, it is about a
thief who steals information by
getting into people's dreams.

It received five Academy Awards.

It starred Leonardo DiCaprio and
Marion Cotillard among others.

It made more than $950 million in
revenue at the box office.
Correct statements in green, manually added
wrong statements in red.

✔ Grounded in Fact [1]


✔ Grounded in Fact [4]


✖ Not grounded

✔ Grounded in fact [2]

✖ Grounded

Overall grounding score:
54%

44. Proprietary + Confidential
Demo : Vertex AI Agent Builder