Intelligent Process Automation with Jetson-TX 2 Murali Kaundinya - PowerPoint PPT Presentation

Intelligent Process Automation with Jetson-TX 2 Murali Kaundinya

Imagine if we could … engage enable end users processes to learn to reflect on a workflow or UX from actionable, real-time that can be improved insights from operational intelligence • and then quickly improve it themselves or recommend and • augment human decision making deliver a proposed solution. with AI and analytics • No different than your experience • to experiment faster with dynamic & participating in OSS project intelligent workflows 2

About me… Murali Kaundinya • With Merck & Co. for 2+ Years • Horizon 3 - Research & Development • Artificial Intelligence • • Data Science • Information Management • Previously at : Goldman Sachs • UnitedHealth Group (Now Optum) • Sun Microsystems (Now Oracle) • Started career at • NASA Goddard Space Flight • Center, Greenbelt, MD 3

Analytics Predictive Prescriptive Augmented Machine Unattended Learning Workflow IPA Process User Experience Invoice NLPs/VPAs Matching Sentiment Email Analysis Analysis Patterns of Work Computer Vision

Use Cases Optical Virtual Data Release Character Customer Unification Engineering Recognition Assistants 5

Architecture 1 3 WF Progressive Web Application 4 6 Pre 2 Embedded BPM Engine Proc WF Post 5 Pluggable ML Layer (TensorFlow, Tesseract) Augmented Analytics 6

OCR/Document Processing • Claims Processing • Invoice Processing • Language Translation • Data Leakage Protection • Reduce Errors/Costs VCAs (aka Chatbots) Continuous Delivery • Release Notes Generation • SOPs/Customer Service • Change Request • Chat Log History • CAB Artifacts • Extracting Intent • SVMs, Naïve Bayes, LSTMs, • ML on Impact Analysis FFNNs • Augmented Analytics • NLP w/ Semantic Ontologies • Approval Workflow • Dynamic workflow 7

Data Unification • 11 source databases • Data refreshes • OMOP Common Data Model • Schema Updates • N X M Data Conversions • Data Cataloguing algos/tools • End-to-End Integration Source: http://OHDSI.org/CDM • Selective SME Involvement Data Similarity Algorithms Analytics Data Visualization Attribute Identification Step 3 Step Trigram cosine similarity. • 2 Step 1 TF-IDF cosine similarity. • Minimum Desc. Length(Jaccard) • Domain Specific Views Welch’s t-test for numerical values. • Common Data Model Entity Consolidation Data De-duplication • Claims OMICs EMR RCT Clustering • 8

FEBRL DataWrangler Snorkel HoloClean 9

Target State Architecture Data Visualization Analytics Step 3 Step 2 Step 1 Domain Specific Views Common Data Model Claims EMR RCT OMICs 10

What’s next? • End-to-End integration • Can GPU databases speed up data unification? • Brytlyt • MapD, Kinetica, Sqream • BlazingDB • Blazegraph – Graph Analytics • Better data visualization with GPUs • MapD 11

Open architecture that can engage enable end users processes to learn • to reflect on a workflow • from actionable, real-time or UX that can be insights from operational improved intelligence • and then quickly • augment human improve it themselves decision making with AI or recommend and and analytics deliver a proposed • to experiment faster solution. with dynamic & • No different than your intelligent workflows experience participating in OSS project 12