A Virtual Assistant Ecosystem for Workflow and Workplace Optimization

“HP handles more than 600 million technical support requests each year and pulls in data from more than 20 million devices. Making it easy for customers to find relevant product usage or support information is key to creating a positive experience. Companies, from all business areas, are producing data, generating information and increasing their consumption of that information. Most of the time, such information is available to customers through web portals and content management tools that are usually designed and implemented in a predefined and static way. Sometimes, there is an option to filter results before generating reports. Rarely, are the reports dynamic – in a way that allows the end user to define what is going to be displayed. During this talk, we’ll deep dive into HP’s architecture and implementation of a Virtual Assistant Ecosystem for workflow and workplace optimization built to provide customers with the information they desire. Our proof of concept leverages some interesting features, such as: – Context-based interactions with the focus set on featuring intent recognition and multi-context dialog management; – Proactive monitoring and action recommendations based on proactive maintenance of the workflow; – Dynamic fulfilment of recognized intentions based on a serverless architecture. The Virtual Assistant Ecosystem is built upon data produced by a wide range of Databricks/Spark services/jobs applying different analytical and deep learning techniques. To offer a complete experience, we are allowing users to connect from different platforms: from Slack channels to mobile applications and chatbot extensions on existing portals to a Virtual Contact Center to support phone calls.”
展开查看详情

1.WIFI SSID:SparkAISummit | Password: UnifiedAnalytics

2.A Virtual Assistant Ecosystem for Workflow and Workplace Optimization Rafael Zotto, HP Franco Vieira, HP #UnifiedAnalytics #SparkAISummit

3.//ABOUT US //RAFAEL ZOTTO Holds a master degree in Computer Science focused on high-performance computing. Specialized in parallel and distributed computing with a special interest in cloud and serverless computing. Works for HP Inc. for more than a decade acting as a software engineer for print firmware and wearable technologies. Currently works in the data science research team as a software engineer and solutions architect having most activities related to applied AI and conversational interfaces. //FRANCO VIEIRA Professional with vast experience in information technology working in software development, cloud computing, and machine learning projects. In the past 5 years, Franco has been developing new services focused on activity recognition, content sharing, and device health. Currently, works in HP Personal Systems Software producing machine learning solutions on the edge. Franco holds a degree in computer science. #UnifiedAnalytics #SparkAISummit 3

4.//LEARNING FROM EXPERIENCES #UnifiedAnalytics #SparkAISummit 4

5.//UNIFIED APPROACH TO ANALYTICS #UnifiedAnalytics #SparkAISummit 5

6.//RICH INSIGHTS #UnifiedAnalytics #SparkAISummit 6

7.//PROBLEM STATEMENT _LOTS of INFORMATION _Just a FEW reports _Not always CONFIGURABLE _TIED to development cycles //Offer a QUICK way to share INFORMATION #UnifiedAnalytics #SparkAISummit 7

8.//CONTEXT //FLEET MANAGER //END-USER _MANAGES a fleet of devices _USES one or more devices _Need to UNDERSTAND status _Wants assets AVAILABLE _FIX problems _Interested on PERFORMANCE #UnifiedAnalytics #SparkAISummit 8

9.//OUR APPROACH //Be AVAILABLE //Offer DYNAMIC ways to access the INFORMATION //Using NATURAL LANGUAGE #UnifiedAnalytics #SparkAISummit 9

10.//INVESTIGATION //Generate STRUCTURED QUERIES from Natural Language _SEQ2SQL, SQLNet, Coarse2Fine, … //INTENT recognition / Chatbot PLATFORMS _DialogFlow, LUIS, AWS LEX, Watson, … #UnifiedAnalytics #SparkAISummit 10

11.//CHATBOT PLATFORM ANATOMY COMPOSED MIGHT INTENTS BY UTTERANCES HAVE SLOTS FULFILLMENT //EXAMPLE _“I need to travel from San Francisco to Philadelphia next weekend” SLOT 1: Origin SLOT 2: Destination SLOT 3: Date #UnifiedAnalytics #SparkAISummit 11

12.//CHATBOT AS PART OF A SOLUTION //Our INSIGHTS and PREDICTIONS refers to a DOMAIN //Our DOMAINS can be mapped to INTENTS _Battery _Thermal _CPU _… //INTENTS can be fulfilled with INSIGHTS and PREDICTIONS _MULTI-CONTEXT dialog management #UnifiedAnalytics #SparkAISummit 12

13.//HIGH-LEVEL OVERVIEW WEB PORTAL SLACK FULFILLMENT INTENT LAMBDA _DYNAMIC response created CALL CENTER RECOGNITION FUNCTION DESKTOP APP USES ANALYTICS SERVICES MOBILE APP SAVES TO QUERY CONTEXT is created for the SESSION GRAPHQL RDS #UnifiedAnalytics #SparkAISummit 13

14.//DEMO //EASY access to our GOLD DATA //UNDERSTAND user needs. #UnifiedAnalytics #SparkAISummit 14

15.//CREATING A PLATFORM WEB PORTAL _Support for SLACK MULTIPLE INTENTS FULFILLMENT INTENT LAMBDA CALL CENTER RECOGNITION FUNCTION DESKTOP APP USES ANALYTICS SERVICES MOBILE APP SAVES TO QUERY GRAPHQL RDS _Gathering data from MULTIPLE SOURCES #UnifiedAnalytics #SparkAISummit 15

16.//LEARN FROM MISSED UTTERANCES //UNDERSTAND what we still DON’T KNOW INSIGHT DISCOVERY MISSED UTTERANCES INPUT TEXT ENTITIES KEY PHRASES INTENT RECOGNITION FILTERING SLOTS KNOWLEDGE NEEDED //EXAMPLE _“What was the health grade of my device fleet yesterday” KNOWLEDGE : 0.99+ KNOWLEDGE: 0.99+ DATE: 0.98 #UnifiedAnalytics #SparkAISummit 16

17.//ONGOING WORK _DEVICES are THINGS connected to our stack. _We have INSIGHTS and PREDICTIONS ready to be used. _Why not DELIVER them to the interested part? //IMMEDIATE delivery _As soon as detected, NOTIFICATION is delivered to the USER. //SCHEDULED delivery _Kept to be delivered as a FUTURE or RECURRENT NOTIFICATION. #UnifiedAnalytics #SparkAISummit 17

18.//DELIVERING INSIGHTS WRITES TO SIMPLE QUEUE WRITES TO ANALYTICS SERVICES EVENT SINK PRODUCE INITIATE EVENTS PROCESSING DELIVERED AS MQTT TRIGGER NOTIFICATIONS ACTIONS _EXTENSIBLE list of PLUGINS _ACT to address an issue _WATCH the system #UnifiedAnalytics #SparkAISummit 18

19.//DEMO 00 #UnifiedAnalytics #SparkAISummit 19

20.//NEXT STEPS //It’s all about USER-EXPERIENCE _HELP fleet-managers. _AMAZE end-users. _ACT in advance. _UNDERSTAND users profile. #UnifiedAnalytics #SparkAISummit 20

21.DON’T FORGET TO RATE AND REVIEW THE SESSIONS SEARCH SPARK + AI SUMMIT