Continuous accuracy and efficiency of Large Language Models (LLM) is key to successfully building out your next AI-infused automation, regardless of business use case.
For our next Connector Corner webinar, we’ll explore how using a seamless AI integration process provides access to industry leading models, curated activities, and embeddings that help achieve operational efficiency.
Join us on March 26 to learn about:
Accessing large language models, hosted by UiPath
Reducing complexities of prompt-engineering, by using curated sets of activities
Assuring accuracy and safety, by building an AI Trust Layer to moderate the output of AI models, and their generated results.
Discovering what’s new in embeddings connectivity
Cultivating your AI knowledgebase using Vector Databases
Expect to see these use cases in action:
Leveraging UiPath hosted LLMs and activities
Document comparison using our LLM framework
Please stay tuned for additional use cases
Speakers:
Charlie Greenberg, host
George Roth, Technology Evangelist
Scott Schoenberger, Senior Product Manager
Koji Takimoto, Director Product Support
3. 3
Connect to all your automation opportunities
with UiPath Business Automation Platform connectors
Microsoft Entra ID (formerly AAD)
Generative AI Automation
Line of Business
Automation
IT system
Automation
Automate any API-based system
No pre-built connector available for your needs? Easily create no-
code custom connectors with UiPath Connector Builder
[Connector examples by category]
4. 4
Scott Schoenberger
Senior Product Manager,
George Roth
Technology Evangelist
Diana Gray
Host, Community Marketing
Manager
Charlie Greenberg
Host, Product Marketing
Manager
Today’s UiPath Team
6. 6
Bing BanG of AI – October 2022
Visual representation combines Big
Bang with launch of Chat GPT in
October 2022.
This artwork creatively merges the
cosmic inception of the universe with
the symbolic milestone in AI
technology, showcasing a
metaphorical 'big bang' of technology.
Dall-E
7. 7
It’s a wonderful life ( the movie )
Bedford Falls becomes Pottersville: The town is much seedier and
more depraved without George's influence.
Mary Hatch, George's wife, is unmarried: Mary, who is George's wife
in his real life, is shown to be an old maid, working at the library.
Without George, she never married or found happiness.
George's brother, Harry, died as a child: Harry fell through the ice
and drowned because George wasn't there to save him. As a result,
Harry wasn't able to become a war hero who saved the lives of
many soldiers during World War II, because he died in childhood.
Many people's lives are worse off: Characters whom George had
helped throughout his life are shown to be in dire circumstances
without his intervention. For example, without George, the druggist
Mr. Gower accidentally poisons a child because George wasn't
there to prevent the mistake.
The pivotal part of the movie involves an angel named Clarence
Odbody, who is sent from Heaven to help George.
Clarence shows George what the world would have been like if
he had never existed:
8. 8
• Slower innovation in software development tools
• Reduced efficiency in debugging/testing
• Limited AI in creative industries
• Impact on custom software solutions
• Slower progress in automating content creation
• Reduced public discourse on AI ethics and regulation (Europe,
US)
• Chatbots and customer service
• Delayed enhancements in accessibility technologies
• Impact on education and learning
• Influence on scientific research
• Impact on Competitive Dynamics
• Delayed Impact on Employment and Skill Demand
• New job types: prompt engineers, explainability engineers, sales
bot manager
If generative AI was not launched in 2022
It’s a wonderful life (without AI)
13. 13
UiPath Business Automation Platform
--and AI
“Most firms do not currently use Gen AI,
Microsoft’s Copilot or other such tools in
a systematic way, even if individual
employees play around with them.” The
Economist
UiPath provides the means to use AI in a
systematic way across the enterprise
through Automation.
14. 14
What is needed to deal with a mesh—
of Gen AI resources ?
The need for
Integration Service and
connectors to AI technologies
16. 16
Secure
UiPath: our GenAI North Star
Technology
agnostic
• Address data privacy concerns
• Maintain content moderation
• Manage hallucination prone responses
• Support customer technology stacks
• Empower RPA developers to take full
advantage of AI opportunities, across their
automation journeys
Use case
driven
• Expand user ability to build and
configure enterprise-grade automations
• Provide verified ‘prompts’ with easy-to-use
connectors and activities
17. 17
Open and Enterprise Ready GenAI
in your Automations
Generative AI Connectors Models
OpenAI GPT-3, GPT-3.5, GPT-4, GPT-4V
Azure Microsoft OpenAI GPT-3 (+embeddings), GPT-3.5, GPT-4
Amazon SageMaker
Falcon (Hugging Face*), Llama2 (Meta*) any other custom AWS model AWS
SageMaker is in Public Preview
Amazon Bedrock Titan, Claude2 (Anthropic*), Jurassic (AI21*), Llama2 (Meta*)
Google Vertex PaLM2, Unicorn, Gemini, Gemini Vision
IBM WatsonX
Granite, Llama2 (Meta*),
Built by IBM using UiPath Connector Builder (GA – all Cloud users)
Anthropic Claude2, Claude Instant
Connector Builder Build your own AI-based connector
*vendor
18. 18
• Upsert Embeddings: Allows you to add data to the Pinecone
database.
• Query Embeddings: Lets you easily search for similar data within
Pinecone for use in text and chat generation.
• Create Embedding: Changes text into a form that can be added to or
searched in a database.
• Generate Chat Completion: Now equipped with database
connectivity to create AI responses that are aware of proprietary
context using an existing knowledge base and semantic similarity.
Azure Open AI
Connector (GA)
Submit prompts with context
‘Grounding’ GenerativeAI Outputs
Insert prompts with context to generate outputs with proprietary information
Pinecone
Connector (GA)
Store vectors for use by LLMs
21. 21
Introducing 'UiPath GenAI Activities'
- First party Integration Service Connector offering access to UiPath managed, popular LLM's
- Pre-built prompts allowing users to focus on automating common tasks with confidence
- Context-enabled, removing technical barriers to referencing large sets of proprietary data
- AI Trust Layer governance and auditability
22. 22
Customer
data & apps
UiPath AI Trust Layer
PII & sensitive data filtering
Usage telemetry and audit dashboards
Gen AI feature policies
LLM
Gateway
No
data
retention
or
training
Third party
LLMs
Context grounding
UiPath LLMs
New!
New!
23. 23
Upcoming curated Gen AI Activities
UiPath Roadmap, Subject to Change
Activity Context
Enabled
Use Cases
Rewrite / Email Generation Yes
Generate written content for internal/external audiences that matches a particular
tone
Match internal brand guidelines
Categorize Yes
Identify key characteristics of information to effectively route through other
automations
Support tickets to correct teams
Format Change No
Change data between popular formats for use in subsequent activities
CSV > JSON
Language Detection No
Identify languages
Route inquiries to the appropriate team based on language
Translate Yes
Detect and translate languages
Translate emails and customer feedback, generate custom responses and
translate back to original language
Named Entity Recognition Yes
Define and identify specific categories and values within documents and data
Extract and categorize all relevant data from medical studies for use in
early- stage drug research
Analyze Image Yes
Generate unique descriptions and answer custom questions about an image
Describe a product inventory based on images of products
24. 24
Context grounding in GenAI Activities
24.4 Preview User Action 24.4 Preview
Capabilities
Upload Data: User will need to upload data into
UiPath Orchestrator buckets
Supported data types
include PDFs, JSON,
CSV
R.A.G.
Activities
Make data LLM-Ready:
Context Grounding will
index and convert data in
specified Orchestrator
Bucket to embeddings
• Predefined
ingestion/R.A.G.
search
parameters
• Works to
complete an
end- to-end flow
with Text and
Chat Completion
Activities
R.A.G.: Search and
retrieve context from
index, send to LLM as
enhanced instruction set
for prediction generation
Context Service: Improve the quality and value of Gen AI in
UiPath Gen AI experiences by grounding LLM prompts with
contextual evidence from UiPath and Customer-Managed
knowledgebases.
Roadmap: These activities will become core parameters to support First
Party AI Activities with R.A.G.
Context Service delivers a retrieval augmented generation (R.A.G.) architecture to
produce more accurate, reliable, domain-specific GenAI predictions
25. 25
Context grounding scenarios:
- Reference a large file in an automation
- Persist the file to a knowledge base (Orchestrator bucket)
- Reference a pre-existing knowledge base (Orchestrator bucket)