6. Payment Fraud
• Compromised Payment Instruments (e.g., stolen cards)
• Intentional Non-Payment (e.g., pre-paid cards)
Account Takeover/Compromise
• Username/Password
• API Key
Abuse
• Free Tier Misuse
• Premium Phone Number
Fraud comes in all shapes and forms
7. Business Rules vs Machine Learning
Business Rules look for specific conditions or behaviors
• Business Rules are easily explained and validated
• Sample New Account Registration rule:
ML Models learn more general patterns by looking at lots of examples
• When fraudsters make small tweaks, the model still recognizes them as
suspicious since it’s unlike anything it has seen from legitimate
customers
• ML models are not just good at finding the risky patterns, they’re much
less brittle than rules
If IP_ADDRESS_LOCATION == [’Japan’] and CUST_ADDRESS_COUNTRY == [‘JAPAN’] and CUSTOMER_PHONE_LOC ==
[‘Spain’] THEN Investigate
Prevention Detection
8. Fraud detection with ML is also difficult
Top data scientists are
costly & hard to find
One-size-fits-all
models underperform
Often need to
supplement data
Data transformation +
feature engineering
Fraud imbalance =
needle in a haystack
9. Introducing Amazon Fraud Detector
A fraud detection service that makes it easy
for businesses to use machine learning to
detect online fraud in real-time, at scale.
13. Example template: Online Fraud Insights
• Detect risky events based on an event’s attributes
• Best for detecting potential fraud when historical
account/user data is limited
• Inspired by models and techniques used to protect AWS
account registration
• Use cases: new account, first transaction, guest checkout
• Inputs: 3 required data elements and 50+ optional
14. Benefits of Amazon Fraud Detector
• Build high quality fraud detection ML models faster
• Stop bad actors at the door
• Built-in online fraud expertise
• Give fraud teams more control
16. Current Customer Issue
1.5B+ hours medical audio data generated per year.
Valuable insights are ”trapped” in the audio.
90%+
generated
by clinicians
Even more billions of hours of medical audio backlog.
18. Unlocking insights in medical audio
Amazon Transcribe Medical Amazon Comprehend Medical
19.
20. Amazon Transcribe Medical
What is it?
Amazon Transcribe Medical is an automatic speech
recognition (ASR) service that enables developers to add
medical speech-to-text capabilities to their voice-enabled
applications.
22. Amazon Transcribe Medical
How do I use it?
Amazon Transcribe Medical
1) Call the API
”Patient is 42 year old male with…”
3) Get a stream of text
2) Pass an audio stream
Device Agnostic
27. Introducing Amazon CodeGuru
• Machine learning service for automated code review and application
performance profiling
• Trained on decades of knowledge and experience at Amazon
• Evolves with user feedback
• Searches for optimizations continuously, even in production
• Provides actionable recommendations to fix identified issues
• Automatically inspects code for hard to find defects
• Helps you find the most promising methods for optimization in your
running application
It is like having a distinguished engineer on call 24x7
28. Amazon CodeGuru: Using ML to Code Review and Optimize High-
Performing Applications
Easily identify performance
and cost improvements in
production environment
CodeGuru Profiler
Detect and optimize
the expensive lines
of code pre-prod
Built-in code reviews
with actionable
recommendations
CodeGuru Reviewer
31. Code Review Key Challenges
• Expertise
• Availability, Compliance and Correctness aspects often do not get addressed because of
lack of expertise.
• Senior Talent
• Code reviews often demand a senior engineer to be involved. Teams may not have the
right individuals or they may be focused on other high value tasks.
• Multiple functional areas
• The number of topics which require expertise, e.g., AWS API use and concurrency, is
increasing
• Human code reviews often focus on business logic and less on
functional correctness.
• Number and size of source code repos increasing
• Reviews often require inspecting a large amount of source code for context
32. Amazon CodeGuru Reviewer
Flags critical defects and reliability issues in source code.
Amazon CodeGuru Reviewer augments human code review
process and does not replace it
Pull
Request
Approval
Merge
Code
Review
Branch
Make
changes
locally
Amazon CodeGuru Reviewer
33. Code Areas addressed by CodeGuru Reviewer
AWS Best Practices: Correct use of AWS APIs
Incorrect use results in performance (e.g., polling) or correctness and completeness (e.g., pagination)
issues.
Concurrency: Correct implementation of concurrency constructs.
Incorrect use results in correctness (e.g., missing synchronization) or performance issues (e.g.,
excessive synchronization) and hence impact availability.
Resource Leaks: Correct resource handling
Incorrect handling (e.g., not releasing database connection) results in slowdown and impacts
availability.
Sensitive Information Leak: Leakage of Personally Identifiable Information
Leakage of sensitive information (e.g., logging of credit card number) leads to compliance issues.
Code defects discovered by mining data: Hard to find defects
Correcting issues (e.g., not creating a client for each lambda invocation) improves code quality.
38. Lynn’s main challenges with poor application performance
• Poor end-user
experience
• Rise of performance problems:
Troubleshooting of distributed applications is
challenging
• Not enough performance engineers: Scarcity
of performance engineering expertise
• Higher cost of
compute
infrastructure
• Impact on mission
critical systems
Business impact Causes
• Losing customers
• Performance optimization is challenging:
not a domain expertise for most developers
39. points directly to
the performance
problem
provides
actionable
recommendations
helps you prioritize
work on code fixes
continuously
learns
performance
best practices
What if we have someone on-call who…
has performance
engineering
domain expertise
continuously
analyzes system
performance
40. CodeGuru Profiler finds your most expensive lines of code
• Trained to find methods with high-potential for performance
optimization
• High latency & low throughput
• High CPU utilization
• Recommends how to fix your code
• Intelligent profiler trained by many years of performance engineering
experience at Amazon
41. Built for production systems
• Low overhead
• Continuously runs on production
• Continuously analyzes performance
• Currently supports applications written in Java
51. Contact Lens for Amazon Connect
Advanced search Detailed analytics &
sentiment analysis
Automated contact
categorization
Theme detection
(coming soon)
Supervisor assist
(coming soon)
Open and flexible
data
Contact Center Analytics for Amazon Connect powered by Machine Learning
The out-of-the-box experience makes it easy for contact centers and their staff to use the
power of ML with just a few clicks.
52. Full text search and sentiment analysis
Conduct fast, full-text search on
the Amazon Connect contact
search page
Search for contacts based on
keywords in transcripts, speaker-
separated sentiment scores (-5 to
+5), and other call characteristics
such as non-talk time resulting from
significant pauses in conversation.
Contact Lens for
Amazon Connect
53. Allen Smith
AI Powered Insights & Analytics
Transcript of voice and chat
interactions with in-line
sentiment markers
Interactions automatically organized
by your defined categories
Contact Lens for
Amazon Connect
Quickly visualize the customer
experience