Understandable Artificial Intelligence

Infrastructure Review

Building On Principals, A New Frontier

Cloud Resources

EDS Data Center

Combined Computing Power

Client Interface

Scala

Zookeeper

Asynchronous
- Actions occuring at the same time, in any order, without waiting on each other
- Current solutions are sequence oriented
Distributed
- Connecting many smaller systems to work together
- Current solutions grow by sequencing faster
Modular
- This principal means components can be easily repurposed
- As opposed to current monolithic designs

Scalable and Redundant
- Built expecting failure and can easily 'drop-in' new resources live
- Current solutions would need to be shut down and migrated
Analysis Focused not Storage Focused
- We store data multiple times, in multiple forms, focused on insight
- Current solutions minimize storage cost, but make insight costly
Flexible Data Consumption
- Takes all data comers, SQL, NoSQL -- structure, unstructure
- Current solutions have strict data structure requirements

Cassandra	Database	Meets All	Puts Data of Any Kind Closer To Analysis For Speed and Flexibility
Hadoop	Distributed Disk	Meets All	Quickly Consumes Any Type of Data
Mesos	Scheduler	Meets All	Coordinates Tools for 'Drop-in' Flexibility
Zookeeper	Failover Coordinator	Meets All	Coordinates Redistribution of 'Duties' During Failures Or When Connections Are Dropped
Spark, Scala, and R	Analytics	Combined -- Meets All	When Combined, Tools Provide Analytics In Ways That Align With Principals
Play	Front-End	Meets All	Visualizes to client and collects client information

Component	Purpose	Principal	What It Does

Point to point encryption over the network
- Data is never exposed without encryption
Client to server encryption
- In line with best practices at banks. Each client session is encrypted regardless of where the client is connecting from
Cloud compute not cloud storage
- We use the cloud for compute power not storage, we control the physical location of each data point
Multi-tiered security layers
- Designed with multiple layers of security so each must be compromised separately

Scale and speed
- Add nearly an unlimited number of 'factors' - qualitative/quantitative
- Can run in seconds/minutes and so could be done 'live' for clients
- Each analysis is applied for all
Data communication - goodbye to silos
- Each part of the business process can share information
More in depth client feedback
- Better understanding of client interest