View
9
Download
0
Category
Preview:
Citation preview
Curious Bot
Antonio Savasta – Matteo Finistauri
AI&RPA Lead Manager – AI Tech Architect and Team leader
VIRTUAL AGENT
Market-leading companies use Artificial Intelligence (AI) to build their Digital Workforce
Integration Service (RPA and Application Integrators)
Automated
execution of routine
processes
Robotic
Process
Automation
“No touch”
approach
Automated
processing of
unstructured
documents
Natural
Language
Processing
Document and
communication
classification
Automated
customer
interactions and
internal support
Digital
Assistants
Automating
routine front-end
interactions
Automated learning
Machine
Learning
Automated and
semi-automated
data analysis and
insight mining
Smarter decision
making and KPIs
Expert
Systems and
Analytics
Operational
decision
optimization
Automated
processing of
images and visual
inputs
Computer
Vision
Analyzing and
classifying Images
Company
Bot Plattform
IT Bot
• Helpdesk
• Tickets
• Q&A
• WLAN Access
• Orders
Product Bot
• Product information
• Service questions
• Q&A
Sales Bot
• Assistant
• CRM Infos
• Voice Frontend
• Product Information
Service Bot
• Sparepart Info
• Product Info
• Voice Frontend
• Time Booking
Compliance Bot
• QM
• Org Info
Knowledge Bot
• Voice Frontend
• Topic Information
• Search Integration
….. B
ot
Ad
vis
ory
Tech
no
log
y B
riefi
ng
, B
usi
ness
Im
pact
& C
ase
s
Based on CuriousBot
Delivery Model (Planning, Support, Marketing, Training, …)
Change Management (Communication, Training, Culture)
Ru
n B
ot‘
s(R
un
an
dle
arn
org
an
isati
on
)
Bot StrategyThe Bot Strategy
creates the base for a
solid development of a
voice portal to your
company and
processes. It shows the
way in which thematic
areas are integrated
one after the other.
Chatbots are divided
into a continuous
landscape and ensure
thereby a solid
operability of the
solution.
Avanade sees chatbots as part of a sustainable bot strategy
Digital Agent
Azure Web App or Bot Service
High Level Architecture of a Bot based on Microsoft Bot Framework
…
Telegram
Messenger
GroupMe
Skype
Slack
SMS
Web
Additional Channels
Node js or .Net Chatbot Web Application
Bot Connector
EndPoint
Bot Builder SDK
Cognitive API’s and
AI Services
Speech VisionLanguage
Machine
Learning
Cognitive
Services
Knowledge Search
Backend Systems and Services Authentication
MS Teams
Avanade
CuriousBot
Bot Service can be deployed
in any Cloud or on-premises
Alternative Services
can be integrated,
e.g. DeepL
Translation
CuriousBot as accelerator for chatbot implementation
Spell Check
Reporting
Configurable LUIS Integration Multi-lingual
Prepared for Machine Learning
Optimizeddecision ranking
RPA preparedEmotion
Input channels
Cognitive Services
Recommendations
Knowledge Exploration
Entity Linking
Academic Knowledge
QnA Maker
Custom Decision
Bing Autosuggest
Bing Image Search
Bing News Search
Bing Video Search
Bing Web Search
Bing Custom Search
Bing Entity Search
Translator Text
Bing Spell Check
LUIS
Web Language Model
Text Analytics
Linguistic Analysis
Bing Speech
Translator Speech
Speaker Recognition
Custom Speech
Computer Vision
Face
Content Moderator
Emotion
Video
Custom Vision
Video Indexer
Dialogue Layer
Perception Layer
How do CuriousBot and Microsoft Cognitive Services work together Machine Learning & Knowledge Base Layer
Bing Speech
Text Analytics
QnA Maker LUIS
TranslationChat
History
Crawler Process
Knowledge Bases
Decision Logic
Response Generator
Bing Speech Translation
pass to Bing Search results
Other actions:
request to ticket system
pass to Self Service Portal
Model Training
Dialog data flow
Emotion data
Avanade´s CuriousBot further enhances Microsoft´s bot framework´s potential in three additional ways
Dialogue Configurability Intelligent Decision Making On-Premises Infrastructure
Dialogues within a bot solely based on Microsoft
Bot Framework must be created, altered and
deleted directly in the source code.
There is no direct configuration capability for
the dialogue flow.
By default the Microsoft Bot Framework is not
offering intelligent decision when using cognitive
services. It provides building blocks, but not a
ready-to-use solution.
Combination of multiple Cognitive Services such
as LUIS and QnAMaker requires programming
and advanced experience.
The standard approach of Microsoft Bots is
completely cloud-based with the option of a
hybrid setup for the core bot logic.
A pure on-premises solution is not possible out-
of-the box. Custom programming is required to
keep sensitive bot dialogues on-premises, and
use Cognitive Services only in an anonymized
manner.
Our Management Application provides user-friendly bot configuration
Editing dialogue steps, adding and editing QnAs, adding utterances and training LUIS models can be done within our central management application which is easily accessible via the browser.
Example of a graphical Chatbot integration (Screenshots)
User Interface
“Ask Anna“ is just one example
of the elegant and smart
integration of a chat bot into
any website.
Together with our UX
department, a group of user
guidance and design
professionals from the agency
sector, we support customers in
developing further concepts
and implementations.
Generalist SpecialistIT Knowledge
Fast SlowSolution Speed
Low HighSupport Costs
High LowUser Satisfaction
Chatbot supported
self helpService Desk (1st Level) 2nd Level Support 3rd/4th Level Support
Use case, use case, use case – Do’s
• Realistic and impactful for the first bot
• Ideally, a task that every employee needs everyday
• Real KPIs against them with real timelines for ROI
• It has to be simpler, easier & faster with the bot than any other option
Use case, use case, use case – Dont’s
• Throwing a bot at any problem won’t solve it successfully
• Is Conversation the right approach for the use case?
• Solve a real problem, don’t implement a bot for the sake of having one
AZURE CLOUD
Why hybrid topology for CuriousBot?
Easy access to internal backend services
Single https connection from Azure to Bot Service
Easy authentication and control to backend systems
Local storage for anonymized chat history
No Cloud data storage required
Easy to convince Corporate Security
Backend SystemsCuriousBot
Chatbot central logic,
Chatbot landing page
Reverse Proxy
Bot Framework Connector
Connect to Chatbot
SQL
Firewall
User channel
Microsoft Cognitive Services
QnA Maker
LUIS
Text Analytics
Translation
Bing Speech
Emotion
Bing Search
DMZ & INTRANET
Frontend & Backend
Firewall
https / JSON
https / JSON
https / JSON
Concurrent LUIS and
QnAMaker analysis
Dialog Start
Welcome CardMO Greeting
User Input
Check FAQ(QnAMaker)
LUIS AnalysisLUIS Recognized
Answered?Problem
classification
SSP Request
TicketRequest
Show Best Responses
(carousel card)
Self Service Portal
request
Ticket request
Open Ticket
PostQuestion?
Core NWS
No
Write NWS and Confirm
Open Ticket and Confirm
Yes
Yes
Collect Feedback
No
Start Bing Search
Show Bing Search results (carousel card)
Answered?
Yes
NoYes
Yes
Nothing Recognized
Explain current scope
Cancel
I am learning<help info>
The chatbot leverages multiple knowledge sources
Managed Content
MO Productive QnA
MO QualityQnA
Content Management
App
O365 General
Questions
New Work Style Social
Support Forums
API
Crawler
Manual Import
User Interaction
A crawler continuously collects new knowledge
MO Production QnA
Quality CheckQnA
Approve/Modify via Management
App
New Work Style
Forums
Crawler Process
Upload
• Runs hourly• Reads only answered questions (V1)• Complete crawl
Metadata required:• Source• Modified timestamp• Forum• Answered flag• Approved flag• Synced flag• Answer-ID
Modify
Train/Publish
Manual tasks via Management App:• Check content• Format check• Approve
View filtered by:• Topic• New changes• Very old entries
to be reviewed
Post question
on social
support forum
‘Crowdsource
’ Knowledge
Don’t have an answer!
Hey MO!
What is…?
The answer
is: 42
Awesome!
Content management
Automatically
filled by Crawler
process
Used by MO bot
(QnAMaker)
Move. Edit
and Approve
Content
Recommended