The FDA Office of Regulatory Affairs (ORA) manages the process whereby all products imported into United States are screened by electronic systems and human inspections, https://www.fda.gov/ForIndustry/ImportProgram/. About 40 million products are monitored annually resulting in 6 billion data records that need to be processed every night. Booz Allen built an Apache Spark system to analyze the FDA ORA data and to predict violations. The solution uses enterprise friendly SQL framework to expand from data aggregation to Machine Learning without heavy coding. The system enables any enterprise DBA or analyst easily access, filter and transform data to apply the latest machine learning models. These analysts are able to process 6 billion records from various databases and other sources every night without any prior experience with Apache Spark. This helped to scale the Apache Spark solution enable data warehouse/RDBM experts to process powerful analytics workloads without needing to know Scala or Python.

1. www.boozallen.comJune 5, 2018
Spark + AI Helps
the FDA Protect
the Nation
Kun Ei Kang
Booz | Allen | Hamilton
Chief Innovation Architect
BOOZ|ALLEN|HAMILTON
Jonathan Chu
Booz | Allen | Hamilton
Chief Technologist

2. www.boozallen.com
AGENDA
2
4
2 3
1

3. www.boozallen.com
FDA AND YOU
3
§ The FDA oversees products which account for 20
cents of every dollar spent by consumers
§ Foreign production of FDA-regulated goods and
materials has exploded over the last decade
§ FDA-regulated food and medical products originate
from more than:
– 150 countries
– 130,000 importers
– 300,000 foreign facilities

4. www.boozallen.com
FDA IMPORTS PROGRAM
4
Responsibilities:
§ Protect public health by electronically screening
all FDA regulated product imports
§ Determine and stop product line/shipments if
they pose significant risk
§ Obtain samples of products for further
laboratory screening as needed
High Volume
~40 Billion shipments in 2017
with continuous annual increase
of 5-10%
Performance Insight
Sheer volume of data led to
increased labor to assess
performance of electronic
screening rules
Change Management
Need for more robust capabilities to
evaluate proposed changes to
screening criteria for importsChallenges

5. Innovation center, Washington, D.C.

6. www.boozallen.com
ODP TURNS DISPARATE DATA INTO VALUE ADDED SOLUTION FOR
OUR CLIENTS
COLLECT
Get and track
data from the
source
PROCESS
Give Data the
Power of
Greater
Context
AGGREGATE
From disparate
data sources,
one version of
the truth
EXPOSE
Abstract away
complexities
through a
single interface
VISUALIZATIONS
APPLICATIONS
DATA SCIENCE
ANALYTICS
BUSINESS INTELLIGENCE
CURATION & DISCOVERY
DATA CONSUMERS
Security, Governance, Provenance, Lineage
Insights
DATA SOURCES
SOCIAL MEDIA
NEWSFEEDS
WEB CRAWLERS
PROPRIETARY
SOURCES

7. www.boozallen.com
ODP CORE TECHNOLOGY STACK
§ Best-of-breed open source technologies chosen, configured and
integrated
§ Containerized data ingest and processing pipeline Apache Spark
and Docker
§ Automated deployment into Amazon Web Services (AWS) and
Azure
§ Ability to swap in-and-out technologies based on use case, as well
as tailor deployments based on use case (e.g. Elasticsearch vs
Solr; deploy search capabilities without Hadoop ecosystem)
§ Data Management dashboard for job management, data tracking
lineage/provenance, metadata management, and governance
§ Analytics platform that democratizes data science and enables
analytic decision making for anyone (of any skill set) in an
organization
ODP Framework

8. www.boozallen.com
SOLUTION JOURNEY
PERFORMANCE ANALYTICS
Provide analytics and dashboards
for assessing current import
screening rules performance
SCREENING OPTIMIZATION
Improve imports screening
through Machine Learning + AI
AGGREGATION
Preparation of data for FDA
subject matter experts
8

9. www.boozallen.com
PHASE 1: AGGREGATION
CHALLENGE
Billions of records in various
database tables dating back to
2007
Spark jobs migrate data into data lake
Filter and join data set
Redistribute data for performance
9

10. www.boozallen.com
Spark Submit
ODP IMPLEMENTATION AT FDA
Spark SQL Job.jar
15 Spark Packages
Connectors
Spark Cluster
Spark SQL WEB UI
JSON
Connection config
SQL

11. www.boozallen.com
PHASE 2: PERFORMANCE ANALYTICS
Leverage Client’s Tableau Investment
Create Tableau Dashboards
Enable Parquet to Tableau TDE format
CHALLENGE
With thousands of rules in place,
provide intuitive user experience
and insight to provide clarity on
current rules performance
11

12. www.boozallen.com
TABLEAU CONNECTOR

13. www.boozallen.com
PHASE 3: CURRENT PHASE - BUSINESS OPTIMIZATION
Recommend changes to model
Determine impact to organization
Determine projected performance
GOALS
Automate use of data captured
during imports entry review and
other data to ascertain risk of
products to improve entry review
process
13

14. www.boozallen.com
DEMONSTRATION OF ODP
14

15. www.boozallen.com
WHAT WE LEARNED
§ The use of Spark further facilitates innovation due to its ability to quickly experiment and
demonstrate business value
§ Due to regulatory mandates, there will always be a human factor in the final decision
“Information is the oil of the 21st century,
and analytics is the combustion engine.”
– Peter Sondergaard -
15

16. www.boozallen.com
CONTACT US
For more information about Open Data Platform, contact:
Jonathan Chu
chu_jonathan@bah.com
opendataplatform@bah.com
https://boozallen.github.io/opendataplatform/
Kun Ei Kang
kang_kun@bah.com
16

17. www.boozallen.com
ODP REFERENCE SLIDES
17

18. Innovation center, Washington, D.C.
An Open Source data management platform harvested from our Booz Allen developer community

19. www.boozallen.com
19This document is confidential and intended solely for the client to whom it is addressed.
2020
130
exabytes
40,000+
exabytes
x2
2005
~40%
“touched” by cloud
computing providers
~33%
of the digital universe will
be valuable if analyzed
DATA CHALLENGES CONTINUE TO GROW EXPONENTIALLY
SOURCES: IDC – The Digital Universe in 2020, The
Forrester Wave: Big Data Hadoop Solutions, Gartner
2016 Metadata Management MQ, Gartner Newsroom
DIGITAL UNIVERSE
DATA GROWTH
(by 2020)
(by 2020)
x300
5,200+ GB
per person
Forrester estimates that companies are only making use of 12% of the data they own
GROWTH OF SENSITIVE DATA
REQUIRING PROTECTION
< 1/3
2010
> 40%
2020
of which < 50%
is protected
unprotected data will
grow by a factor of 26
GROWING NEED OF
DATA GOVERNANCE
Gartner estimates data lakes will still not have
effective metadata management until at least 2018
By 2020, information governance and policies will
be managed by metadata alone
Poor quality of data costs an average organization
$13.5 million per year

20. www.boozallen.com
THE TRUTH ABOUT BIG DATA
20
“The more data I have, the more questions I can answer, the more
knowledgeable my organization will be”
It’s about the right data, combined with the right tools,
for the right problem— not how much data is stored
BELIEF REALITY
“Through 2017, 60% of big data projects will fail to go beyond piloting and experimentation, and will be abandoned.”
Gartner, http://www.gartner.com/newsroom/id/3130017
Policies, restrictions, and security need to continue to be enforced
and the original origin of the data and any changes to it need to
be tracked throughout the life of the data asset
Successful organizations leverage industry technologies,
but they also adopt an agile culture that enables them to
try new ideas and pivot quickly based on lessons learned
Stored data needs to be cataloged to enable analysts to
exploit the data. Information such as data set name, data
formats, data tagging, release-ability, retention, and other
metadata must be available.
“If I combine all my data together I will have full insight
into what my organization is trying to accomplish”
“If I leverage industry technologies, I will be successful”
”Once I have all my data in one place, I can focus on analytics”
vs.

21. www.boozallen.com
WE NEED TO EMPOWER ORGANIZATIONS TO MOVE
FROM DATA CHAOS TO POWERFUL ANALYTICS THAT
DRIVE INSIGHT INTO THEIR MISSION
21This document is confidential and intended solely for the client to whom it is addressed.

22. www.boozallen.com
OPEN DATA PLATFORM CAPABILITES
22
DATA SECURITY
Ensure security is applied and data
is protected from the point it enters
to the platform until it leaves
DATA GOVERNANCE
Gain insight into the health of your
data feeds, users of your data (both
human and system), metadata and
policies, and the lineage of data
across the entire platform
DATA GATEWAY
Access data across your enterprise
using an open common API
DATA INGEST
Fully established and orchestrated
Docker-based ingest pipelines
supporting RDBMS, Web Services,
JSON, HL7, ICPUBS, and more
DATA QUALITY
Refine your data either during or post-
processing to add greater context to
the data and enhance quality
PLATFORM AUTOMATION
Fully automated deployment ODP
enabling repeatable deployments
with no manual intervention
DATA ANALYTICS
Derive powerful insights quickly
through readily available analytical
capabilities

23. www.boozallen.com
ODP TURNS DISPARATE DATA INTO ONE VERSION OF
THE TRUTH
COLLECT
Get and track
data from the
source
PROCESS
Give Data the
Power of
Greater
Context
AGGREGATE
From disparate
data sources,
one version of
the truth
EXPOSE
Abstract away
complexities
through a
single interface
VISUALIZATIONS
APPLICATIONS
DATA SCIENCE
ANALYTICS
BUSINESS INTELLIGENCE
CURATION & DISCOVERY
DATA CONSUMERS
Security, Governance, Provenance, Lineage
Insights
DATA SOURCES
SOCIAL MEDIA
NEWSFEEDS
WEB CRAWLERS
PROPRIETARY
SOURCES

24. www.boozallen.com
EXCHANGE – FIND & CURATE
The power to take control of your data
INTEGRATED TEAMS & APPLICATIONS: Dissolve technical and cultural
barriers by building a community for all skill levels. ODP
empowers both technical staff and non-technical staff to get
value from data.
FILE TYPE AGNOSTIC: Links directly to any URL or data file,
anywhere, using our intuitive Web extension.
CUSTOMIZED CURATION: Organizes data sets on clients’ terms.
Collects data sources into custom “nets,” then curates them
by topic.
DATA-GATHERING COMMUNITY: Users can create a profile, curate
data, and gain followers. A social data community levels out
the playing field to find quality data.
CROWD-SOURCED QUALITY: The Comment, Like, and Star Rate
features give everyone a say about the quality of data
sources.
CUSTOM SOLUTION PACKAGES: One size does not fit all. Every
organization faces unique challenges, and no one should
settle for a standard, shrink-wrapped, stand-alone tool.

25. www.boozallen.com
EXPLORE – ANALYZE & DISCOVER
25
Data science & machine learning for the masses
EMBRACE MODERN ANALYTICS: Gain greater data insights through
machine learning, natural language processing, advanced data
querying, data curation bookmarking.
GROW A DATA SCIENCE CAPABILITY: More than just tools, ODP helps
build a data science community and nurtures the growth of an
analytic capability.
API TO ANALYTICS: Downloads data locally with one click, or sends data
via API to Sailfish Explore.
NATURAL LANGUAGE INTERFACE: Users can ask “plain-English” questions
about the data—no special query language is required.
VISUAL QUERY BUILDER: Users can simply drag-and-drop to build
complex queries, without coding.
WORKFLOW MANAGEMENT: Users can think through their analytical
approach in real time, saving their workflow, sharing it with others,
and/or scheduling it ahead of time.

26. www.boozallen.com
ODP ARCHITECTURE AND TECHNOLOGY
STACK
26Booz Allen Hamilton Internal
Data:
• Streaming
• Batch
• Structured
• Unstructured
Raw Data Zone
• Unaltered Source Data1. Collect 2. Prepare 3. Organize 4. Profile 5. Enrich 6. Aggregate
• Establish data
policies
• Cleanse and
normalize data
• Extract metadata
• Store and
catalog
• Data Archive
• Assess Data Quality
• Manage Version
Control
• Track provenance /
lineage
• Enhance
metadata
• Create
relationships
• Combine disparate
data sources,
create single
version of truth
Trusted Data Zone
• Master Data
• Cleaned, Processed, Organized
• Correlated Entities
• Policy Enforcement
• Version Controlled
Sandbox Data Zone
• Combination of trusted and raw data
• Data Sampling to support Development,
Experimentation, and Prototypes
Analytic Data Zone
• In-Memory
• Optimized Indexes
• Graph Analysis
• Data Warehouses
• Machine Learning
Data Sharing Services
• Search
• Query
• Mutation Services
• Enterprise Application Integration
• Publish/Subscribe
• External Data Connections
• Business Process Management
LDAP / ADFS
Data Security
• Cell Level Security
• Row Level Filtering
• Attribute Security
• Centralized Policy Enforcement
Data Governance
• Data Catalog
• Data Quality
• Data Provenance and Lineage

27. www.boozallen.com
TYPICAL IMPLEMENTATION STRATEGY
2-3 Weeks
Proof of Concept
6-8 Weeks
Pilot
Varies by Customer
Operational
• Demonstrate technical
capabilities in Booz Allen
environment
• Target single contextual use
case
• Public data sources
• Demonstrate technical
capabilities in Client sandbox
environment
• Sample client data sets
ingested
• Pilot use case implemented
against client needs
• Operational deployment in
Client environment against full
client needs
• Planning, Installation
• Accreditation
• Client data sets ingested
• Monitoring/management of
Open Data Platform
• Platform optimization based on
operational use
• Upgrades
Ex. For a prospective commercial client, we ingested
publicly available geospatial, health, and financial
information in 3 weeks to showcase our ability to
respond to health events worldwide
Ex. For a government client, we ingested multiple
acquisition data sets to produce example analytics
and enable the client to reduce license costs while
increasing visibility across functions of the enterprise
Ex. For a government client, we support a multi-year
deployment ingesting over 100 feeds to support
strategic analysis of threats against US assets and
provide Defense/Intel level data security

28. www.boozallen.com
WHY LEVERAGE THE ODP?
§ We have packaged our experience and expertise into an automated package to give organizations a “hot-
start” to analyze data quickly
§ We have researched, integrated, and hardened best-of-breed Open Source Software using our experience
gained in Defense, Intel, Civil, and Commercial deployments
§ It is backed by a community of Big Data Professionals, Certified Cloud Architects, and Security
Professionals to ensure successful, low-risk delivery
§ It solves the challenges with deploying and managing a platform, wrangling and managing data feeds, and
integrating best-of-breed Open Source software
ODP enables Booz Allen to partner w/ clients in an Open Architecture

29. www.boozallen.com
A GLIMPSE OF OUR BIG DATA EXPERIENCE
* Defense / Intel Deployments on SIPRNET and JWICS
Plus Commercial Pharmaceutical, Gas &
Oil, and Telecommunications