AWS Summit Series 2018 - San Francisco: Nhung Ho, Head of Data Science for Intuit

Introduction

The speaker introduces herself and the company she works for, Intuit. She talks about the company's mission to power prosperity around the world and how they achieve this through their products.

Introducing Intuit

• The speaker introduces herself and her company, Intuit.

• Intuit has three core products that serve two broad groups of customers: Mint and TurboTax for personal finance management and tax preparation, and QuickBooks for accounting for small businesses and self-employed individuals.

• The company's mission is to power prosperity around the world by making financial management easy for everyone.

Journey to the Cloud

The speaker discusses how Intuit has been moving its infrastructure, applications, data, and machine learning to the cloud since 2013. She explains why elasticity is important for their product and how AWS services help them handle bursty traffic patterns.

Moving to the Cloud

• Since 2013, Intuit has been moving its infrastructure, applications, data, and machine learning to the cloud.

• Elasticity is extremely important for their product because of highly seasonal traffic patterns in tax preparation industry.

• On peak day last year there were 2 million tax returns filed in TurboTax online product alone.

• AWS CI/CD products allow them to quickly deploy changes with minimal impact on customer experience.

Machine Learning at Intuit

The speaker talks about how they are incorporating machine learning into their product experiences. She describes what her team used to have to do to get a machine learning model into production and how their partnership with AWS is helping them scale this out.

Incorporating Machine Learning

• Intuit has been incorporating machine learning and AI into their product experiences in the past few years.

• Before taking any models into production, they have to define who will be using it, how they will use it, and what their SLA requirements are.

• Scaling up from an MVP version of a model to production requires vastly different compute resources.

• Their partnership with AWS is helping them scale out their machine learning capabilities.

Building and Deploying Models with AWS SageMaker

The speaker discusses the challenges of building and deploying machine learning models, and how AWS SageMaker can help overcome these challenges.

Challenges of Model Deployment

• Delivering model results in real-time requires embedding application code within a back-end application or building a microservice.

• Building microservices is not the specialty of data scientists.

• Load testing and performance testing are necessary before production.

Benefits of AWS SageMaker

• Using SageMaker eliminates the need for data scientists to wear multiple hats (data architects, engineers, QA, DevOps).

• Predefined architecture and elastic computer environments make deployment fast and efficient.

• Auto-scaling infrastructure comes out-of-the-box.

• Deployment time is cut down by 90%.

Future Plans with AWS SageMaker

• Leveraging the full power of SageMaker deployment and training with EMR for near real-time model retraining.

• Higher-level services such as AWS Transcribe and Comprehend will allow for greater insights for customers.

• Personalization will start from first sign-on, anticipating what users need next.