UiPath Community Day Kraków: Devs4Devs Conference

Cristina Vidu
Global Manager, Marketing
Community

Mikolaj Zielinski
RPA Consultant
Proservartner
Dawid Kot
Digital Transformation Lead
Kamil Miśko
RPA Senior Developer/
Solution Architect
Zurich Insurance

Agenda
8.30 -Welcomecoffee-30'
9.00 -Opening note/ Intro to UiPath Community -10'
9.10 -Cloud migration -Proservartner& DOVISTA case study -30'
9.40-From bottlenecks to breakthroughs: Citizen Development in action -25'
10.05 -Next-level bots: API integration in UiPath Studio -30'
10.35 -Coffee Break -15'
10.50-Document Understanding with my RPA Companion-45'
11.35-Power up your Robots: GenAIand GPT inREFramework-45'
12.20-Lunch Break –1hr
13.20 -From Concept to Quality: UiPath Test Suite for AI-powered Knowledge
Bots -30'
13.50 -Communications Mining -focus on AI capabilities -30'
14.20-Polish MVP panel: Insights on MVP award achievements and career
profiling-30'
14.50-Networking Hour (drinks & finger food) -1hr

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Dawid Kot
Digital Transformation Lead
Proservartner

aka. Artificial Artificial Intelligence

MARCIN DROZDOWSKI @DOVISTA
PAWEŁ KAMIŃSKI @DOVISTA
MIKOŁAJ ZIELIŃSKI @PROSERVARTNER
UIPATH COMMUNITY DAY KRAKÓW:
DEVS4DEVS CONFERENCE - 27.06.2024
CLOUD MIGRATION

https://wykop.pl/cdn/c3201142/4e5ac560f1164363a5f161dadcd2dd8d91c5fd42870a111bcdca65ffe3839790.jpg?author=Jossarian&auth=992bc5f9ccc9ce6bad2c3ecbd781
313d

Agenda
1.Intro to DOVISTA
2.Intro to Proservartner
3.„The Long and Winding Road” to
the Cloud
4.Why Proservartner?
5.„Migration Operation”:
•Migrationmethodology
•DOVISTA’s use case overview
•Benefits and lessonslearned
6.Q&A

ABOUT US
Marcin Drozdowski
•Automation Center of Excellence
Manager
•17 years of experience in DOVISTA
(Lean, MOST, Configuration, Master
Data)
•Hobby: sci-fi, playing the drums,
cycling
Paweł Kamiński
•RPA Developer / Project Lead
•5 years of experience in DOVISTA
(Sales, Planning)
•Hobby: basketball, computer
games,programming in Visual Basic

20
Mikołaj Zieliński
Aprofessional solution architect with extensive experience in
software and automation development using UiPath
technology. Mymain area of focus is product strategy and
building the roadmap for partners, with a particular emphasis
on system migration and modernization efforts.
My expertise lies in guiding organizations through complex
migration processes, ensuring seamless transitions to newer
platforms while minimizing disruptions and maximizing
operational efficiency. With a deep understanding of legacy
systems and modern architectures, I have successfully led
numerous migration initiatives, enabling companies to
leverage the latest technologies and unlock new capabilities.
My strategic vision and technical acumen have been
instrumental in helping UiPath customers future-proof their
automation investments and stay ahead of the curve in an
ever-evolving digital landscape.

WE ARE
A UNION OF DISTINCTIVE BRANDS
AND ONE UNITED TEAM.
WE ARE
DAYLIGHT PIONEERS.
WE ARE
WINDOW AND DOOR PEOPLE.
WE ARE

DOVISTA is part
of the VKR Group
The VKR Group consists
of companies within two
separate business areas.
Roof Windows & SkylightsVertical Windows & Exterior Doors
VKR Group companies
together employ about
17,200 employees in
over 40 countries.

DOVISTA IS A UNION OF DISTINCTIVE BRANDS WITH STRONG LOCAL ROOTS
23

DOVISTA´s history
24
1941
V. KannRasmussen & Co is
established in
Copenhagen, specialising
in glass roofs
1962
VillumKann
Rasmussen
registers the
trademark VELFAC
2000
SP Fönster
(Sweden) and
TrarydFönster
(Sweden) join the
Group
1999
Rationel Vinduer
(Sdr. Felding)
joins the Group
2004
DOVISTA is
established
2002
Mockfjärds
(Sweden) joins
the Group
2006
Natre
(Norway)
joins the
DOVISTA
Group
2005
OH Industrijoins
the DOVISTA
Group
2012
Lian
(Norway)
joins the
DOVISTA
Group
2017
New strategy
2016
KRONE
joins the
DOVISTA
Group
2021
Slovaktual, Wertbau,
Dobroplast, Egokiefer
and WEBCOM join
the DOVISTA Group
2021
WERU GmbH
(Germany) and Unilux
GmbH (Germany) join
the DOVISTA Group

Ourunionofdistinctive
brandshasmorethan
750 YEARS
ofexpertisewithdeveloping
andprovidingvertical
windows andexteriordoors.

“We bring daylight and fresh air
into people’s everyday lives.”
ALLAN LINDHARD JØRGENSEN
CEO & PRESIDENT

27

28
WE DEVELOP AND
PRODUCE VERTICAL
WINDOWS AND
EXTERIOR DOORS MADE
FROM:
WOOD,
WOOD/ALUMINIUM , PVC,
PVC/ALUMINIUM,
AND ALUMINIUM

WE OFFER UNIQUE VERTICAL WINDOW AND EXTERIOR DOOR SOLUTIONS
TAILORED TO OUR CUSTOMERS’ LOCAL NEEDS AND PREFERENCES
OUTWARD
SASH/FRAME
INWARD
SASH/FRAME
LIFT/SLIDING
DOORS
FACADE
DOORS

WE ARE MORE THAN
6,000 COLLEAGUES

WINDOW VILLAGE I
Facts & figures:
•Over 1100employees
•39-average employees’ age
•41% -share of women
•522 670 elements produced in 2021
•6 077loaded trucks left WVI in 2022 (117
weekly)
•7,5 Mmeters of wood used yearly (=2x
the length of Polish borders)
•7,2 Mmeters of aluminum used yearly
•956 030 m² of glass used in 2022 (=area
of 134full-size football fields)

T5
Multicolor/
T8
T4
T7
T6
T3
T2
T1

DOVISTA AUTOMATION COE FACTS
113 automated processes
(E.g. Outbound & Inbound Logistics, Sales Support,
Finance, Capacity Planning, Service)
4 RPA
Developers
1 Business
Analyst
10 on-prem
prod.
machines
2 RPA Controllers
(Role embedded in IT
Service Delivery)

OUR
CASE STUDY

”But do we really have to do it…?”
•Challenging start →Long ramp-up
due to:
•Temporary lack of internal
resources
•Other business priorities -some
business-critical automations to be
implemented first
•Lack of specialized knowledge -
'once in a lifetime' project area
•Fear of opening ”Pandora's box”-
unknown and/or very old processes
https://tenor.com/pl/view/do-not-touch-it-programmer-walking-cow-coding-gif-17252607

Start with Why
•Top 3 main reasons:
•To reduce technical debt
•To merge separate setups after
acquisitions
•To gain quick access to the
latest UiPath tools & features
(Jumpstart AI adoption)

Choosing a Partner
Why a partner?
•Risk mitigation (”Insurance policy”) -We needed to
ensure that processes would be migrated smoothly,
without stopping the running robots that business
operations rely on.
Why Proservartner?
•Local market presence, same language, time zone,etc.
•Personal brand and previous experience with UiPath
cloud migration of Mikołaj Zieliński.
•Previous successful collaboration with Dawid Kot.
•Responsiveness (”Big enough to deliver… small enough
to care.”)

Migration Methodology

Kick-off:
20.11.2023
Audit &
Planning:
December
2023
Consolidation:
January 2024
Cloud
Migration:
February-
March 2024
Hyper-care:
April 2024
Celebration:
10.04.2024
Project Phases @DOVISTA

Pre-Migration Audit: Key Highlights & Benefits
▪Comprehensive assessment of existing RPA setup
(orchestrators, robots, processes, infrastructure)
▪Identification of business-critical automations to
ensure continuity during migration
▪Informed decision-making on migration approach,
resource allocation, and architectural design
▪Opportunity for license optimization and cost savings
▪Evaluation of compliance, security, and data privacy
requirements
▪Alignment with industry best practices and future-
proofing the RPA setup
▪Minimizes disruptions to critical processes
▪Enables data-driven migration planning
▪Facilitates license and cost optimization
▪Addresses compliance and security concerns
▪Allows for future-ready architecture design
▪Lays the groundwork for successful RPA
consolidation and cloud migration
BenefitsKey Highlights

Comprehensive
Assessment of Existing
Setup
01
The audit involves a thorough
evaluation of the current RPA
setup, including the orchestrators,
robots, processes, and
infrastructure. This assessment
helps identify potential challenges,
dependencies, and areas that
require special attention during
the migration.
Pre-Migration Audit
Minimizing Disruptions
to Critical Processes
02
Informed Decision-
Making
03
License Optimization
04
Compliance and Security
Considerations
05
Future-Proofing the RPA
Setup
06
By carefully examining the
existing processes and their
dependencies, the audit
enables the identification of
business-critical
automations that must
remain operational
throughout the migration.
This information is crucial
for developing a migration
strategy that minimizes
disruptions and ensures
business continuity.
The audit provides valuable
insights into the current
state of the RPA
environment, enabling
informed decision-making
regarding the migration
approach, resource
allocation, and architectural
design. This data-driven
approach helps mitigate
risks and ensures a well-
planned and efficient
migration process.
During the audit, the team
can assess the current
license utilization and
identify opportunities for
optimization. By
consolidating licenses and
aligning with the latest
licensing models,
organizations can
potentially achieve
significant cost savings.
The audit allows for the
identification of any
compliance or security
concerns that need to be
addressed during the
migration. This includes
assessing data privacy
requirements, regulatory
constraints, and
implementing appropriate
security measures in the
new cloud environment.
The audit provides an
opportunity to evaluate the
existing RPA setup against
industry best practices and
emerging technologies. This
insight enables the team to
design a future-ready
architecture that can
accommodate growth,
scalability, and the integration
of advanced capabilities like
AI and Machine Learning.

Migration Waves
Wave 1:
20
processes
2 weeks
Wave 2:
23
processes
1 week
Wave 3:
20
processes
1 week
Wave 4:
20
processes
1 week
Wave 5:
14
processes
2 weeks
Wave 6:
7 processes
1 week

Additional Benefits
•Processes and setup
optimization / clean up:
•New naming convention
•Proper roles assignments
•Conversion from Windows legacy
to Windows compatibility
•Migration from Classic Folders to
Modern Folders
•Scalable setup and folder
structure in the orchestrator
•Well-documented setup and best
practices
•MS Graph training and
implementation

On-premises
Two separate Orchestrators:
Increased complexity and
inefficiency due to separate
Orchestrators for New and Old
DOVISTA
Three Access Links:
Cumbersome management and
potential errors due to three
different links to access robots
Different Admin Credentials:
Added complexity and potential
security risks due to different
credentials for the admin
account
UiPath
Automation Cloud
Single Orchestrator
Three environments on One
link:Access to three
environments via a single link,
encompassing both existing and
new versions of DOVISTA,
facilitating testing and transition
between different system
versions
Easy access to admin panel:
Convenient access to the
administrative panel, simplifying
system management and
monitoring without complex login
or authorization procedures
Simplifying the Setup

Key Success Factors
•On-site kick-off meeting
•No language barrier
•Audit phase and detailed
planning
•Proactiveness
•Fixed time reservation for regular
meetings
•Highly motivated and engaged
Team
•Business communication and
engagement
•Collaboration with Infrastructure

Lessons Learned
•It wasn't as difficult as we
expected, but...
•Engage team membersto the
fullest extent possible, with an
external consultant serving as a
guide and mentor.
•Select a project management
tool at the very beginning of the
project (and stick to it
throughout the entire duration of
the project ☺).
•It is never too early to ask the
business for test data.

Q&A

Contact us:
Marcin Drozdowski
linkedin.com/in/marcindrozdowski
Paweł Kamiński
linkedin.com/in/pawelkaminskii
Mikołaj Zieliński
linkedin.com/in/mikzielinski

49

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Michał Cieślak
Senior Manager, Automation Programs
McCormick & Company

From Bottlenecks to Breakthroughs
Citizen Development in action

Michał Cieślak
Senior Manager
Automation Programs & Platforms at McCormick

McCormick & Company

McCormick & Company
Uniting the World through Flavor

McCormick & Company
Our Brands

Citizen Developer Program a’laMcCormick

Citizen Developer Program in McCormick
How we got here and where we’re heading…
2019
Pilot group of 4CDs
StudioXas sole tech used
2020
Additional 20CDs added to the group
StudioXlicenses provided to everyone
2023
CD Program restarted & rebranded
50CDs added
Tech portfolio expanded
2024
Next 55CDs added to the group
Global participation across
functions & regions
2025+
Ongoing expansion of CD Program
Moving to distributed model

Challenges, Speedbumps, Surprises

Citizen Developer Program in McCormick
What we encountered along the way…
Our findings
•Huge CD potential to fill the gap in automation left by RPA
•Challenges with engagementfrom some CDs
•Significant time requirement to master StudioX
•Procedural bottlenecksin automation development (IT, audit, processes etc.)
•The dread of selectorconfiguration & management
•Value of personal connectionto CD success

Achievements, Successes, Insights

Citizen Developer Program in McCormick
What we achieved and learned in our journey…
Our achievements
•Enterprise-wide recognitionof automation potential
•32 automations built using StudioX
•20 Citizen Developers using UiPath StudioX + 21Attended users
•Automation capability for web portals and systems off limit to bots
Our lessons from CD Program
•UiPath Academy is great, UiPath Academy + own use case is exceptional!
•Volunteeroutperforms Voluntold
•You need space & time for magic to happen
•Personal connection is key to high engagement
•Simplificationis a milestone to success

Citizen Developer Program in McCormick
Use case: Attended automation of pre-deductions in HighRadius (O2C)
Key Info
✓Implemented on Feb 26
th
2024
✓Processing 16k+pre-deductions (p.a.)
✓Estimated690h of time saving (p.a.)
✓Potential expansion to further clients on HighRadius
Problem Statement
HighRadiusisasystemtomanagecashallocationand
deductionmanagement.Whileautomatingsomepartsofthe
process,italsocreatesmuchmanualwork.Pre-deductions’
processforUScustomersishighlymanualandvolume
intensive.Processconsumedmuchtimeandwasexposedto
manualerror.
SolutionDesign
AnattendedautomationcreatedusingUiPathStudioXcanbe
launchedondemandbyARteamtoprocesspre-deductions.
AllitrequiresisthebasicinformationinExcelformat.Thanks
tousingaPiP(PictureinPicture)functionality,automationcan
runwhilstallowingARAnalysttoworkonothertasks.

Thank you!

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Mikolaj Zielinski
RPA Consultant
Proservartner

NEXT-LEVEL BOTS:
API INTEGRATION IN UIPATH STUDIO
MikolajZielinski

What areAPIs?

What is an API?
The API , an Application Programming Interface is a set of rules and protocols that
allowsdifferent software applications to communicate and interact with each other.

Types of APIs

SimplifiedAPIcommunication model
Response
Request
Application 1
Application 2

Response formats

HTML
HTML is rarely used for API responses but can be useful for APIs that need to return web pages or
formatted text.

XML
XML is another common format, especially in enterprise environments. It is more verbose than JSON but
offers more flexibility in terms of data representation.

YAML
YAML is less common but is used in some APIs for its human-readable format. It is often used in
configuration files.

JSON
The most widely used format for API responses due to its simplicity and readability. It is lightweight and
easy to parse, making it ideal for web applications.

Protocol Buffers
Protocol Buffers (Protobuf) is a binary format developed by Google. It is more efficient than JSON and XML
in terms of size and speed, making it suitable for high-performance applications.

Working with JSON
Get data from JSON:
•With JSON.Net •With UiPath Activity – Deserialize JSON
In – JSON string
Out - JObject

APIs inUiPath Studio

API with Integration Service
1.Go to Integration Service and create new
Connection.
2.In studio make sure that you are working in
same folder where you created integration.
3.From activities in tab Available find your
connector and pick your action. You can
automate with connection which you created
in cloud service.

Send API via http Request in Studio
UsecURLor fillform.

Send API via http Request in Studio-payload as Text file
1.Save example request payload in text file.
2.Read Text file in studio
3. Perform Deserialization string to Jobject–its good practice as its validating
the payload format.
4.If you performed step 3 add JsonObject.ToStringas Body in HTTP Request
activity. If not, use the string which you got from Read TextFile.
5.In request propertiesadd:
-Request URL
-Set Request Method
-Set Accept format to JSON

Send API via http Request in Studio-payload as JObject
If we want to send API with Paylodgenerated from
JObjectyou have to build Object based on call schema.
1.Initialize new JObject.
2.Add fields –every filed is JProperty, it contains
name of filed and its value. If schema contains
SubObjectyou can:
-Build new JObjectas variable and use it
within another JObject.
-Define another JObjectwithin
current JObject
3. In request propertiesadd:
-Request URL
-Set Request Method
-Set Accept format to JSON

Send API via .netRestSharpin Studio-payload as JObject
1.Initialize Rest Request object.
2. Initialize Rest Client object.
3.Initialize JObjectwith payload data as you did in
previous example.
4.Integrate payload with your rest request object.
5.Execute call and fetch data from it.

Which method is faster?
•API with Integration Service –700ms
•API via http Request in Studio –cURLimport -352ms
•API via http Request in Studio –text file as source -1671ms
•API via http Request in Studio –JObjectas source –1263 ms
•Sending API with RestSharp.Net–1483 ms

Which method is best?
•RestSharp: advanced method which allow to customize all details of API
sender.
•cURLwith UiPath HTTP Sender: fast method, but you are missing the source of
schema of call.
•Text file with UiPath HTTP Sender: good for sending always same calls, if file
will miss or get corrupted it won’t work anymore.
•JObjectwith UiPath HTTP Sender: stable and fast design, it require some skills
to build it.
•Integration Service: the easiest way to consume API

Thank You

Bold transformation. Real results.
Delivered with assurance, speed andimpact.
We partner with clients to design and implement next generation operating
models, processes and automation.
Proservartner brings experienced practitioners, data, tools and independence to
unlockclient opportunities.

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Ewa Gruszka
Enterprise Sales Specialist AI & ML, SEE
UiPath

Document Understanding
•The UiPath word mark, logos, and robots are
registered trademarks owned by UiPath, Inc. and its
affiliates.
©2023 UiPath. All rights reserved.
with my RPA Companion

Get your documents
processed intelligently
Teach your robots to understand documents
using AI-enhanced skillsfor data extraction
and interpretation.
Drag and drop these capabilities directly into
your automation workflows to embed AI

Built for: RPA and citizen developers –no data science skills required
Built-in AI: Specialized AI models and Generative AI
Built as: SaaS, APIs, self-hosted or air-gapped integrated with UiPath Business Automation Platform
1
Receive 3
Act2
Understand
Documents
•Multiple languages
•Various types & formats
•Handwriting & signatures
•Tables & checkboxes
•Skewed & low-quality scans
AI-powered processing Human in the loop UiPath Automation
•Initiate other automations
•Input into system of record
•Drive other actions
Validate the
extracted information
and handle exceptions
DigitizeClassify Extract

Move-For-You Co.
We move so you don’t need to move
PO: NP74006735
1 February 2020
PAYABLE WITHIN 15 DAYS OF RECEIPT
20800 ALMADEN AVE, SUITE 404
SAN JOSE, CA 95120-0520
T: +1 425 555 9876
F: +1 425 555 3456
E: [email protected]
www.moveforyou-co.com
Bill To:
Tony Tzeng
12345 Mango Lane
Seattle, WA 98108
INVOICE DETAILS
Packing services
Storage fees (1 month)
House move (white-glove service)
Vehicle storage and transport
Sales tax 10%
Total Fee including Tax
FEE
$1,282.00
$1,884.00
$5,320.00
$5,186.00
$1,367.20
$15,039.20
Methods of payment
Personal Check: Move-For-You LLC
Wire Transfer: BigBankCo., Account 123456789-0987ABC
Invoice No: 456200-TZE1
Digitize text in the documents
using OCR

Logic in the process to automatically determine if field(s) extracted
correctly
Character
validation
Basic math
Date validation
External source
verification
Invoice # = 7 digits
Total = Line Items + Tax
Valid date format
Verify against PO Record
PO
Number
Vendor ID Vendor
Name
5928452 12345 PROTECH
3758292 98734 LORANA
Extract data from the documents

Generative Extractor
Mostly unstructured
documents
Data extraction –from rules to AI
to a hybrid approach
RegEx-Based
Extractor
Structured fields
Machine Learning
Extractor
Mostly semi-structured
documents
Intelligent keyword
Classifier
A combination of both –
rule-based and AI-based approaches
Mostly documents combining both structured and less structured formats
Rule-based AI-based
Hybrid approach
Note: Forms AI will be soon replaced with the next-gen Document Understanding experience designed for building Specialized AI models for different document types

Validate classification of the
documents
Classification Station is
used to check, correct, and
confirm the results of
document classification and
splitting.

How Document
Understanding works
Digitize
Classify
Train ML Models
Extract
1
2 3
Structured
Documents
Semi-Structured
Documents
Unstructured
Documents
RPA
BPM
API
Other Systems
Export
5
Digitize
1Structured
Documents
Semi-Structured
Documents
Unstructured
Documents
2
Validate
4
Sample
Production
Confidence Scores < 0.8
Models
- Bilet 0.05
- Kontrakt 0.9
Traveler: Ewa Gruszka Date:
25.09.2023
Class: 2
City from: Warszawa
Kontraktor: ABC Consulting 0.90
Zleceniodawca: Szpital Miejski 0.89
kwota: 5000 0.95
Data: 12.01.2024 0.75

Gen AI Control

Human – Bot Cooperation
Scores control
Business Process
Including data input
OCR, Classification,
Extraction
ML model quality Rules check
Note: Forms AI will be soon replaced with the next-gen Document Understanding experience designed for building Specialized AI models for different document types

CATEGORIZE, DIGITIZE, and EXTRACT
DATA
UiPath Business Automation Platform offersmultiple AI capabilities
to meet the challenges enterprises are facing related to Records
Management and Document Processing at scale
Monitor folders, systems, and
emails mailboxes
Confidence threshold-based
human Validation (Optional)
Automate downstream actions such
as data entry, validation, completion,
etc.
Classify the document,
digitize, and extract
information
Government Agencies are burdened with paper and
communication-based processes, which are often very
essential as part of business process completion
Identify documents based on
configurable retention, archive,
and purge policies
Apply archival, redaction, or
purge policy
Update document management
systems
Classify and tag the
document based on content
Significant volume of records that need to be managed for
retention, purge, and archival to comply with legislative
mandates as well as reduce infrastructure costs.
REDACTION and PII FILTERING
Retrieve the required documents
Confidence threshold-based
human Validation (Optional)
Release redacted dataApply business rules for what
needs to be redacted
Compliance based mandates to redact data and significant
burden on agencies to release data for FOI requests.
Typical Records Management Challenges that organisations
face
How can UiPath’s Business Automation Platform help?
RETENTION, PURGE, ARCHIVAL

Power Up YOUR
ROBOTS with AI
Use of GenAIto build and
maintain you robots
-REFramework
-AiPath

About me
•Krzysztof Karaszewski
•RPA Training Manager - Symphony Ventures
•RPA and LAT Automation Product Manager - HSBC
•Open-source projects:
•REFramework for Power Automate Desktop
•GPT-Powered REFramework
•AiPath
•Microbots AI

Inspiration
•I would like the presented solutions to
serve as an inspiration for utilizing
GenAIin RPA processes.
•These are not ready-made recipes for
quick production deployment.
•The proposed solutions share the same
compromises as other LLM-based
solutions, such as error margins and
incompatibility with the current
architecture.

THE IDEA BEHIND
THE SOLUTIONS
GenAI'sAbility to Identify
UI Elements and
Understand Processes

Solving RPA Challenge
using AI
•RPAChallenge.com
•The aim for AI was to get generate correct
UiPath selector for every input field.
•This is difficult as each time form is submited
the order of UI elements and theirs properties
are generated randomly.
•AI is a brain, UiPath (or anyother automation
tool or solution) are the hands.
•Solution
•Getsimplified HTML code.
•Send it with base prompt to the AI model.
•Get list of input boxes selectors in UiPath format.
•Fill the web form

Comparing the results
•All the models where able to fill the web form.
•Not all of them were about to do it every time.
•Mainly wrong JSON output format.
•Force JSON was not always working.
•Bigger models were not much better than some
small models.
•While being way slower (from 50% to 150% )
•Smaller models usually required lager prompt
with more detailed explanation what to do.
•RPA tools fills it under 5 seconds. Humans 200
seonds
•Claude 3 Haiku a clear winner.
•GenAIhandles identifying elements on a
webpage quite well based on the HTML code of
the website.Model
Result
(seconds)
Reliability
GTP 4 Turbo 103.9 High
GPT 4o 47.7 High
GPT 3.5 57.2 High
Groq LLaMMa3 70B 48.6 Low
Groq LLaMMa3 8B 45.4 Low
Groq Gemma7B 39.2 Low
Claude 3 Haiku 37.2 High
Claude 3 Sonet 46.1 High
Perplexity Sonar Small36.5Medium
Groq Mixtral 8x7B 33.4 High

GPT-POWERED
REFRAMEWORK

Quick VIDEO
https://www.youtube.com/shorts/4m23jQvZBlI

GPT-Powered REFramework
•Integration between UiPath RPA and GPT models that allows to maintain
automation easier and faster.
•Fix not working selectors on the fly without stopping the process
•The automation can run without any predefined selectors at all.
•Ui elements can be defined in plain language.
•Selectors will be generated AI in the background without human intervention
•MakesRPA solution more reliable.
•Even if the target application changes significantly
•It's especially useful if underlying applications are getting frequent up-dates
that changes theirs UI structure or where there are significant discrepancies
between development, test and production applications.
•In such situation RPA developers would not need to fix the solution on
production, but thanks to GPT-Powered-REFramework AI will do it for them
automatically.

GPT-Powered REFramework Architecture
Initialization
Get
Transaction
Data
Process Transacation
Get Config from
JSON
Get selectors from
Application Model
JSON
Component A
Component B
Save selector in the
Application Model
JSON
Get simplified
website HTML
code
Use GPT AI model to
generate a new selector.
Validate selector
generated by GPT.
End Proces

Application Model JSON Structure
•{
•"Ticketing System - RPA Credit Bank": //More than one application can be stored in the file
• "applicationName": "Ticketing System - RPA Credit Bank", //Application Name//Top selector as it would be difficult for
GenAI to generate it. If this selectors is wrong probably the issue is more seveare.
• "applicationHeader": "<html app='chrome.exe' title='Ticketing System - RPA Credit Bank*' />",
• "path": "https://futureofworkblog.web.app/ticketingSystem/SubmitCase_OldVersion.html", //URL or Path to the
application
• "Name text box": {
• "uiElementName": "Name text box", // Name of the UI element based on which it will be recognised.
• "componentName": "1.1 Insert Ticket", // Component that the Ui element belongs.
• "selector": "<html app='chrome.exe' title='Ticketing System - RPA Credit Bank*' /><webctrl id='name' class='validate'
type='text' />„ // Selector
• }
•}

Why JSON?
•Easier to store in the Orchestrator as an asset.
•Has to store information about description, selector,
component and UI element name in one object.
•Easier to operate.
•Convertible to object or dictionary.
•Glob_JT_ApplicationModelvariable to store the information
about selectors.
•Accessible from all related components.
•Getting selector for the object:
•Glob_JT_ApplicationModel("Ticketing System -RPA Credit Bank")("Name text
box")("selector").ToString
•Recommend to name activities using the following format.
•Type Into -Name text box
•Can be converetedto DataTablebut I do not recommend it.

Process Transaction
•Each component is wrapped in Do-While loop
that is retried each time a UiSelectorexception
is encountered. int_maxRetryvariable defines
how many times maximum the component will
be retried before throwing SE.
•Framework checks if it's selector related issue.
•GetType(UiPath.core.SelectorNotFoundException).E
quals(exception.GetType) or
GetType(UiPath.Core.ElementOperationException).
Equals(exception.GetType)
•In_Bool_ProductionRun= TRUE the framework
works like normal REFramework.

Fix Application Model With GPT
•Read Propmpts
•promptTemplate.txt
•conversationTemplate.txt
•GetAndClearHTMLCode
•Get attibute innerHTML activity
•Remove tags like:
•<Style/>
•<Script/>
•<Comment/>
•data-uipath_custom_id
•Send to GPT model:
•HTML code
•Prompt
•UI element name from the
Application Model.json
•Verification if the selector is
correct
•Element Exists
•If element name ends with value
•Get text and check if the obtained
value meets requirements
•RegExvalidation
•GPT validation (not implemented)

PROMPT
•Think step by step and take your time to generate a UiPath selector for the UI element: '{0}'
•Selector needs to be in UiPath format, for example: <webctrl[attribute name]='[attribute value]' [attribute name]='[attribute value]'
... /> and it consists of all the attributes for the given UI element.
•List of rules to follow while creating the selector:
•-For 'class' attribute, leave only the first class name and replace the rest with a wildcard. '*'. ex: [class='classnamenext_classname']
to [class='classname*']
•-For tag button or btnadd 'aaname' attribute to the selector.
•-Selector cannot contain innertextor parentidattribute.
•-Selector needs to start with "<webctrl" and end with "/>"
•-Selector needs to have more than one attribute.
•Provide the result of the final step as JSON ex.: {{[UI element name]: [selector]}}.
•Do not include any additional comments or remarks provide only requested JSON.
•Use the HTML code below to create selectors:
•{1}

If element doesn’t exist or obtained value is
incorrect
•Continue the conversation with AI
•Send information that selector is incorrect.
•Element doesn’t exist
•Or the value is incorrect
•Value of the incorrect selector, to avoid AI clearing the same selector.
•Application specific guidelines if applicable.
•If it is known that selectors usually changes in a specific way in a specific
application, such information can be added to the prompt to make it easier for the AI
to generate a correct selector.
•Using List of Dictionaries to maintain the conversation
•NewDictionary(OfString, String) from
{{"role","user"},{"content",str_conversationContent}}

AI generated selector or SELECTORs
•By default, all selectors for the given component are generated.
•Single action selectors are just to complicated.
•They are saved back to the Application Model JSON
•You can decide, if you want to use new selectors, just save or just
generate
•Very useful for testing and trouble shooting.

How it work in Practice?
•Old „Ticketing System”
•https://futureofworkblog.web.a
pp/ticketingSystem/SubmitCa
se_OldVersion.html
•Old style looks and feel.
•ID for each UI element.
•No JS framework, plain HTML.
•Simple to automate.
•New „Ticketing System”
•https://futureofworkblog.web.a
pp/ticketingSystem/SubmitCa
se_NewVersion.html
•Modern looks and feel
•Order of element changed.
•No ID’s assigned to UI
elements
•Modern JS framework
•More difficult to automate.

How much time would you take to generate
Selectors?
•No ID.
•Inner Text contains the whole text in the
input form.
•Anchors?
•Try it at home.
•Both applications are publicly available.
•Interestingly Object Repository
recorder generates incorrect selectors,
so you must do it manually.
•Points labels inside the input box.
•AI generate all selectors in less than a
30 seconds (GPT-3.5) or 20 seconds
(GPT4o)

Video Presentation
•GPT-Powered REFramework
•https://www.youtube.com/watch?v=m0AbiNFfYG0&t=186s
•Generating Selectors for Future Of Work Challenge
•https://www.youtube.com/watch?v=m0AbiNFfYG0&t=225s

Place for the GPT
Power REF in CI/CD
pipeline

Place for the GPT Power REF in CI/CD
pipeline
Development
(GPT Powered REFramework
TestingProduction
UI selector related exception
Fix selectors and store new selectors in
Application Model. JSON
Test the solution using new
Application Model.
Release
Release back on
production, if all test
completed successfully.

Place for the GPT Power REF in CI/CD
pipeline –DEV/TEST vs. Production
Production
UI selector related exception
Fix selectors and store
new selectors in a new
Application Model.
JSON
Process one
case
UI Related
Exception
Step 1
Step 2
Step 3
Re-try
Step 4
Step 5
Fix
selectors
Re-try
•GPT Power REF is
designed to fix
selectors one-by-one
or all of them at once
for the same
component.
•If the component has
a description, then
GenAIcan determine,
if it was performed
property.
•GenAIcan also
determine if values
obtained during the
process meets the
requirements.

Compromises Taken
•Application Model JSON maintains
•Building components with Application Model JSON in mind.
•The selectors can be generated automatically by the solution from the start.
•Difficult to make it compatible with existing RPA solutions.
•Works best with classic activities
•If you have an object repository it might be difficult to integrate GPT-Powered
REFramework with the Modern Experience solutions.
•Never tried it.
•Costs (very high estimates, for complex application, not optimized, 20k
tokens per selector)
•GTP 4o –0.10 USD per selector
•Claude 3 Haiku: 0.003 USD, GPT 3.5 0.005 USD
•For this presented example, 0.01 USD (one cent) for GTP 4o

AI PATH SOLUTION

AI Robot - AiPath Solution
•Fully autonomous AI robot
•Based on similar open-source projects like:
•Natbot.
•Web Voyager.
•Self-Operating Computer Framework.
•Most of the open-source solutions use Google.com as a starting point.
•I wanted the robot to act in business-like environment.
•Access to business-like applications.
•RPA Credit Bank app, ACME System 1, UiBank
•Log-in to application using „Credentials Vault”.
•Execute even high-level task.
•Claude Sonet3.5 is a game changer allowing to execute even complex
tasks.

AiPath Solution Structure
Get HTML code
•Simplify the HTML code.
•Get rid of unnecessary scripts,
comments, tags, classes.
•Less code means less tokens
means better performance.
•Remove empty objects.
•To simplify creation of the
selector each element is
assigned with custom ID.
1
Core of the prompt
•„You will be acting as a user
interacting with a browser to
achieve the following goal:”
•Goal to achieve
•You can execute following actions:
•List available actions including
„Login to”, „Click”, „Type”
2
Add applications’ data
•Data about applications that the
robot can interact with.
•Name of the credentials to obtain
from the Orchestrator.
3

AiPath Solution Structure
-Response
•JSON formatedresponse that contain the following details
•„step” –Free text contains reasoning that stand
behind taking a specific action.
•„action” –Click, Type etc.
•„selector” –Selector UI element to interact with.
•"goal_achieved": false or true to determine if the
goal was reached.
•Based on these details the next action will be executed.
•Till the goal_achievedflag will be true.
•AI needs to determine
•Which application to open.
•Which action to take.
•And, if the goal was achieved or not.
•The whole solution works as a conversation between UiPath that have
eyes and AI that can reason about what UiPath sees.

Where can such solution fit?
•Exception handing
•Fix selectors.
•Validate processes.
•Automation components that has high rate of exceptions.
•Unexpected pop-up windows.
•Notifications.
•Multiple versions.
•Small high-volume processes
•News screening.
•Web page monitoring
•Processes that requires simple reasoning which would be difficult to
build using regular programming.
•Get Email content and classify it.

Exception handing and Resume Process using
AiPath
Error message
UiPath Process
System Exception Take Screen-shot
Add current HTML
or application
structure
Send it to AILet the AI to determine
-What was the error?
-If is possible to resume
work?
-If the MS needs to be
informed?
Take actions accordingly.
Try to resume the work.

Demo
•https://youtu.be/QnPr9w7GIks

PROCESSING
DOCUMENTATION
WITH AI

Analysing Project Documentation using LLMs
•Take a video of a process and ask it be analysed by the Gemini
Flash.
•More of less accurate process map in less than a 30 seconds.
•Take a description of the process and list of available components
in your library to calculate which components are built and the
time it would take to build a new ones.
•Take a component invoke file (copy it to clipboard) and ask AI to
summarize it or provide documentation for it.
•Use documentation to ask AI to build entire processes for you.
•Invoke files only.

SUMMARY

Few words regarding using ChatGPT for
programming
•Personal experience.
•Most of my solutions would not come to live due to lack of time.
•ChatGPT great for prototyping.
•Not always provides the most optimal solution.
•GPT4 and GPT4o way better than GPT 3.5
•Most vivid example
•Convert API payload from AI calls to C# objects which greatly simplifies
messages management.
•30 minutes of work done in less than 5.
•Recommend it to learn programming.

AIPATH Library
•If you are thinking seriously about using LLMs in RPA
•C# library that contains necessary types for generating queries to GPT
models
•OpenAI
•Claude
•Gemini
•Create a query to AI in just one line of code.
•Especially helpful when it comes to sending queries to Gemini.
•Allowed me to build AI solutions much faster.
•New Object with or New Dictionary(of Sting, Object) solutions are much slower.
•https://github.com/futureofworktraining/AiPathLibrary

CONTACT
•Future of Work Channel
•https://www.youtube.com/channel/UC745wfXwnmOvaYFFUYXB1CA/
•LinkedIn
•https://www.linkedin.com/in/k-karaszewski/
•GitHub
•https://github.com/futureofworktraining

Lunch Break1h

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Kamil Miśko
RPA Senior Developer/Solution Architect
Zurich Insurance

What we want to achieve

Test Automation in the AI Pillar
What we are going to test

Test Automation in the AI Pillar
How is reusability and scalability achieved?
Strategy:
Modular reusable component architecture
Test Automation Framework
Execution Template
Object Repository
Input Arguments

UiPath Test Automation Framework
146
Execution
Template 1
Execution
Template 2
Execution
Template 3
Execution Templates
Setup 1
Prepare to run the test
Setup 2
Prepare to run the test
Setup 3
Prepare to run the test
Run Test
Test Case Placeholder
Run Test
Test Case Placeholder
Run Test
Test Case Placeholder
Tear Down
Final actions
Tear Down
Final actions
Tear Down
Final actions

Execution Template
147

Object repository
148
Key Features
Manage from centralized place
List of UI activities
Capture Element wizard
Drag-and-drop elements
Reuse localy or packed as library
Update in one go with UI libraries
Hierarchical structure

Test Results
149

Test Results
150
Test Executions

Test Results
151
Test Cases

Test Results
152
Assertions

Test Results
153
Assertions

Current Workflow & Further Steps
154
Testing Bot Results
Execute & Results
Publis
h
CI/CD Tool
ALM Integration
Create Defect
Version Control System
App Developers
Pull Request
Test Sets & Test Cases
System Under Test
Test Cases Linking

Examples and Numbers
155
•Multiple alphabets tests
•Document recognition
•Image generation and recognition
•21 test sets
•61 automated test cases
•125 objects captured
•25 activities automated
•1 developer assigned to the project

Community Day
Krakow
I am a speaker at
Devs4Devs conference
Tomasz Wierzbicki
Business Analyst
Office Samurai

Employee-driven
active learning
+
Unsupervised
Machine Learning
Automation
Analytics
+

LABELS
ENTITIES METADATA
▪Concepts, themes and
intents
▪E.g. ‘Change of address
request’, ‘Urgent’, ‘Status
update request’, etc.
▪Organised in a hierarchical
structure
▪Should not be used to
capture information that is
present in the metadata
▪Structured data points
extracted from the text
▪E.g. Policy numbers, trade
IDs, URLs, dates, monetary
quantities, etc.
▪Flat hierarchy
▪Additional structured information
associated with each message
▪Metadata properties can be:
✓User properties (defined and added pre-
upload, e.g. NPS score)
✓Email properties (captured from emails,
e.g. sender, recipients, domains, etc.)
✓Thread properties (automatically
derived by the platform for threaded data
like emails and chats, e.g. # of messages
in thread, thread duration, etc.)
1 2 3

Application
Application > Adobe
Application > Adobe > Photoshop
Application > Adobe > Reader
Application > Browser
QoS > Chaser
QoS > Escalation
QoS > Urgent
Label examples:

Classic automation AI-Powered automation

•Data can be uploaded into communications mining
via out of the box integrations, the UI or via theAPI.
•Current Integrations:
•Microsoft Exchange
•Salesforce
•CSV upload
•More to come
•Via API:
•Data Ingestion Framework (UiPath)
•CLI

We’re here if you have any further questions
Q&A

Cristina Vidu
Global Manager, Marketing
Community

Hello, we’re
UiPathCommunity.
Connect with experts and peers on the
latesttrends in automation and AI.

•Vibrant ecosystem of more than 2.8 million professionalsand students that succeed in
their automation journey. You get to engage in:
Developer eventsand
hackathons
Content creation: blogs,
videos, usecases
Recognition: MVPs and
champions
Forum support & product
engagement
Learning on Academy &
gettingcertified
Start your automation journey here!

•Platform that helps
professionals to get started
with free Automation and
RPA education.
•Hub to interact with experts,
get communitysupport, and
get UiPath product and
technology updates.
•Main hub to participate to local
and global meetups,
hackathons and to register to
community programs.
Main Platforms forEngagement
https://www.uipath.com/rpa/academy https://community.uipath.com/
https://forum.uipath.com/

Users meet and exchange knowledge at community
events around the world
Meetups
Hackathons
Global and local
user groups who
gather online and
offline
Competitions to build
innovative solutions
and drive automation
adoption
500+ events per year
1000+ participants in
hackathons
Blogs/
Use case
repository
Learn from real use
case scenarios and
best practices
400+ use cases and
100+ blogs
Advocacy
Get recognized as
Most Valuable
Professionals and
Community
Champions
Exclusive group of
350+ advocates

Our Community
members
365 days in the Community
1.7MAcademy learners
170,000 Forummembers
100,000 events participants
400+
Automation Champions
125 Most Valuable
Professionals (MVPs)
•552 events
•56+ active countries
•25,000+ topics on Forum
•45 community blogs
•attendees from over
6000+ companies

UiPath Community MVPs are hand-picked experts from our community of users who are the most active,
engaged, and knowledgeable about the UiPath products, and they contribute to making our community better
with each step.
Scan to view our global
2024 UiPath
Community
MVPs

UiPath Chapter Leaders
Organize:
Plan and facilitate the UiPath Community Meetups for the city you live
in on a regular basis. Work with the Regional Community Manager to
bring your community on the UiPath map, while leading and growing a
likemindedexpert's group.
Representation:
Become the face of our local UiPath Community, while gathering
other members around the local UiPath identity and facilitating
networking and knowledge exchange.
Benefits:
-UiPath swag bundle/voucher for you and your local community
-Direct access to the UiPath Community team
-Listed profile on the UiPath Champions page
-Participate in an exclusive group across EMEA and
theworld,dedicated to UiPath Chapter Leaders
Join
UiPath Community Krakow!

Developers and business
users succeeding together
in their automation journey.
Want to learn more?
Let’s talk.
community.uipath.com
[email protected]

Polish MVP panel:
Insights on MVP award
achievements and career profiling

Networking Hour
(drinks & finger food)

UiPath Community Day Kraków: Devs4Devs Conference

About This Presentation

Slide Content

Tags

Categories

Download

Quick Actions

Statistics

Related Slideshows

UiPath Community Day Kraków: Devs4Devs Conference

About This Presentation

Slide Content

Slide 2

Slide 3

Slide 4

Slide 5

Slide 8

Slide 11

Slide 16

Slide 17

Slide 18

Slide 19

Slide 20

Slide 21

Slide 22

Slide 23

Slide 24

Slide 25

Slide 26

Slide 27

Slide 28

Slide 29

Slide 30

Slide 31

Slide 32

Slide 33

Slide 34

Slide 35

Slide 36

Slide 37

Slide 38

Slide 39

Slide 40

Slide 41

Slide 42

Slide 43

Slide 44

Slide 45

Slide 46

Slide 47

Slide 48

Slide 49

Slide 50

Slide 51

Slide 52

Slide 53

Slide 54

Slide 55

Slide 56

Slide 57

Slide 58

Slide 59

Slide 60

Slide 61

Slide 62

Slide 63

Slide 64

Slide 65

Slide 67

Slide 68

Slide 69

Slide 70

Slide 71

Slide 72

Slide 73

Slide 75

Slide 76

Slide 78

Slide 80

Slide 81

Slide 82

Slide 83

Slide 84

Slide 85

Slide 86

Slide 87

Slide 88

Slide 89

Slide 90