Applying a Serverless-First Mindset to bring AI into the Enterprise

AI is becoming increasingly important to every business… but training everyone in your company to be a Data Scientist just isn't practical. Fortunately with AI becoming ever more democratized there are now ways of empowering more and more people within your organization. To do that, let's talk about:
- What a Serverless-First mindset is, and how we can apply it to AI
- What your Software Engineers really need to know about AI to use it
- What AI education non-technical members of the organization might need
- How we can take advantage of things like the AWS AI Stack to:
- Educate Software Engineers without prior AI
- Experience
- Do rapid prototyping and exploration
- Free up your Data Scientists to work on the most differentiating solutions
- Allow Software Engineers to effectively partner with Data Scientists

Transcript

1. Liberty IT Applying a Serverless-First Mindset to bring AI into the Enterprise Gillian Armstrong @virtualgill
2. Liberty IT
3. Liberty IT
4. Liberty IT Gillian Armstrong @virtualgill Solutions Architect, AWS ML Hero Liberty IT, Belfast
5. Gillian Armstrong // @virtualgill Serverless- First Mindset?
6. Gillian Armstrong // @virtualgill An application that costs you nothing to run when no-one is using it (apart from data storage) Serverless Technology is…? Assembling your application using pay per use managed services
7. Focusing on Functionality Offloading everything else Gillian Armstrong // @virtualgill A Serverless Mindset is…
8. Focusing on Functionality Offloading everything else Gillian Armstrong // @virtualgill A Serverless Mindset is… Business Value
9. Gillian Armstrong // @virtualgill Rent (Pay as you Go) Buy Build Software Ethos Only build what you need to / what will be a differentiator for you. Only continue until Value is Realized
10. Gillian Armstrong // @virtualgill How does this apply to AI in the Enterprise…
11. Data Scientists Software Engineers
12. Gillian Armstrong // @virtualgill
13. Gillian Armstrong // @virtualgill AWS ML Stack **other clouds are available
14. Gillian Armstrong // @virtualgill AWS ML Stack Functions Databases Queues Logging Storage API Proxy
15. Gillian Armstrong // @virtualgill AWS ML Stack
16. Gillian Armstrong // @virtualgill - Rapid Delivery of Business Value and - Path to Machine Learning Upskilling across wider group of people Benefits of the AI Services…
17. Gillian Armstrong // @virtualgill - Rapid Experimentation - Demonstrate Art of the Possible (especially in low-context / low data scenarios) - Less up-front need for specialists Rapid Delivery of Business Value
18. Gillian Armstrong // @virtualgill - Low-Cost Prototypes using pay-as-you go services - Lower Maintenance / Management of the AI within your application Rapid Delivery of Business Value
19. Gillian Armstrong // @virtualgill What did you learn? Path to Machine Learning Upskilling across wider group of people
20. Gillian Armstrong // @virtualgill AWS ML Stack
21. Gillian Armstrong // @virtualgill What did you learn?
22. Gillian Armstrong // @virtualgill Why not start at the bottom? Developers love shiny stuff, they love to know how it works… source: am developer But we want to keep a Serverless Mindset, so a focus on Business Value first
23. Employee Digital Assistant Gillian Armstrong // @virtualgill
24. Gillian Armstrong // @virtualgill AWS Content Analysis Solution https://aws.amazon.com/solutions/implementations/aws-content-analysis/
25. Gillian Armstrong // @virtualgill
26. Gillian Armstrong // @virtualgill Deployment Tips… https://aws.amazon.com/solutions/implementations/aws-content-analysis/ - Resources are named based on the name you give the Stack so… - Don’t put ‘aws’ in the name (not allowed for some things like Cognito) - Keep it short or with the generated suffix you’ll go over the 63 chars allowed for an S3 bucket - Be aware some things carry a cost while they are running regardless of whether you are using it (circled)
27. Gillian Armstrong // @virtualgill
28. What did you learn? Possibilities of AI Gillian Armstrong // @virtualgill
29. Gillian Armstrong // @virtualgill Amazon Comprehend Amazon Forecast Amazon Lex Amazon Personalize Amazon Polly Amazon Rekognition Amazon Textract Amazon Transcribe Amazon Translate Amazon Kendra Amazon Fraud Detector Amazon CodeGuru AWS AI Services
30. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Gillian Armstrong // @virtualgill
31. Gillian Armstrong // @virtualgill
32. Gillian Armstrong // @virtualgill
33. Gillian Armstrong // @virtualgill
34. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Gillian Armstrong // @virtualgill
35. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Gillian Armstrong // @virtualgill
36. Gillian Armstrong // @virtualgill
37. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Importance of Data! Gillian Armstrong // @virtualgill
38. DATA • Quantity • Quality • Language / Linguistics • Feature Engineering Gillian Armstrong // @virtualgill
39. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Importance of Data! Applying Software Practices to AI Gillian Armstrong // @virtualgill
40. AI SERVICES Gillian Armstrong // @virtualgill • Coding • Governance / Source Control • Deployment, CI/CD • Testing • Monitor
41. ENGINEERING Gillian Armstrong // @virtualgill AI Everything else • Infrastructure as Code • Deployment, CI/CD • Testing • Security • Observability • Cost
42. Gillian Armstrong // @virtualgill AWS Content Analysis Solution https://aws.amazon.com/solutions/implementations/aws-content-analysis/
43. Gillian Armstrong // @virtualgill For this stack how many… Lambda Functions: Step Functions: 40 30
44. Gillian Armstrong // @virtualgill
45. Gillian Armstrong // @virtualgill For 2 photos and 1 video, how many… Lambda Functions Run: Step Function Transitions: 2,208 2,730
46. ENGINEERING Gillian Armstrong // @virtualgill AI Everything else • Infrastructure as Code • Deployment, CI/CD • Testing • Security • Observability • Cost
47. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Importance of Data! Applying Software Practices to AI Legal, Security, Privacy and Ethics Gillian Armstrong // @virtualgill
48. ETHICS Gillian Armstrong // @virtualgill • Humanizing Technology changes how we relate to it • Automation of Jobs • Privacy Concerns • Human Rights Concerns • Environmental Concerns
49. ETHICS Gillian Armstrong // @virtualgill Always pair “What can we do”, with the question “What shouldn’t we do”
50. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Importance of Data! Applying Software Practices to AI Legal, Security, Privacy and Ethics Changing User Experience Gillian Armstrong // @virtualgill
51. USER EXPERIENCE Gillian Armstrong // @virtualgill • Hey, there are still humans to think about too
52. USER EXPERIENCE Gillian Armstrong // @virtualgill • Mental Models • Explainability and Trust • Feedback and Control
53. Gillian Armstrong // @virtualgill https://pair.withgoogle.com/
54. What did you learn? Possibilities of AI Complexities of AI (expectation levelling) Basic Concepts and Terminology Mapping from Business Requirements/Question to Machine Learning Question Importance of Data! Applying Software Practices to AI Legal, Privacy and Ethics Changing User Experience Gillian Armstrong // @virtualgill
55. Gillian Armstrong // @virtualgill AWS ML Stack
56. Gillian Armstrong // @virtualgill Rent (Pay as you Go) Buy Build AI Ethos Only build what you need to / what will be a differentiator for you. Only continue until Value is Realized.. Move forward based on - Functionality Limits Reached - Cost - Finding a Differentiator
57. Gillian Armstrong // @virtualgill What did you learn? An Understanding of how to evaluate where AI could be used An Understand of how using AI changes your application A move towards a common understanding and collaboration with Data Scientists
58. Gillian Armstrong // @virtualgill Data Scientists Software Engineers How Do I?
59. Gillian Armstrong // @virtualgill Thank you! Get in touch! @virtualgill
60. @Liberty Information Technology @Liberty IT @Liberty_IT