Bring AI and Serverless together and you create a new world in which what you thought you know about software may need to be adjusted. Developers transition from “micro-managers” - telling computers what to do step by step, into “teachers” - assisting computers in learning. The code alone is not enough anymore; if the past decades were spent producing tools to handle the code better, now there’s something more abstract than code, shaped in a human-unfriendly language, called “model”, which is entangled with the code. How does this impact developer experience? Is it easy to manage? Can serverless architectures improve it? This presentation will walk you through a demo AI app built with serverless, composing multiple AI functions into one workflow. The functions will be deployed into a FaaS platform powered by Apache OpenWhisk - the most popular open source serverless platform. You’ll learn about FaaS architectures, open source technologies, as well as areas where serverless streamlines the experience for developers. We'll try to answer the question: is AI development FaaSter with serverless? If you want to learn about emerging technologies enhancing developer experience, or if you’re passionate about AI applications, then this presentation is for you.

1. Making AI FaaSt
.
Dragos D Haut - Principal Engineer @Adobe
Akhilesh Kumar - Applied ML Engineer @Adobe

2. FaaS
Function as A Service

3. a.k.a Serverless

4. FaaS Value Props

5. FaaS Value Props
1. FaaSter to PROTOTYPE services

6. FaaS Value Props
1. FaaSter to CREATE services

7. FaaS Value Props
1. FaaSter to create services
2. Never pay for Idle

8. FaaS Value Props
1. FaaSter to create services
2. Never pay for Idle
3. Low maintenance overhead

9. FaaS: Build more, pay less
1. FaaSter to create services
2. Never pay for Idle
3. Low maintenance overhead

10. ||

11. Evolution of Business Logic

12. Monolith
Microservices
Functions

13. High-level Microservice architecture

14. Multiple regions

15. Microservice Cost & COGS

16. Microservice Cost vs FaaS Cost

17. VS

18. How ?

19. FaaS has better premises

20. FaaS premises
Code - a smaller unit to deploy and scale

21. FaaS premises
Code - a smaller unit to deploy and scale
Request based auto-scaling

22. Making AI FaaSt

23. --- https://twitter.com/andrewyng/status/700086119333892096

25. "With AI, we should look at the
programmer more as a teacher,
rather than a micro-manager. "
— Peter Norvig, Director of Research at Google.

26. " We spent the last 40 years building
up tools to build programs to deal
with text (code) in a good way …"

27. "… but right now we are creating
models instead of text, and we just
don’t have the tools to deal with
that. We need to retool the
industry."
— Peter Norvig, Director of Research at Google.

28. "Neural networks are not just
another classifier, they represent the
beginning of a fundamental shift in
how we write software. They are
Software 2.0."
— (Nov, 2017) - Andrej Karpathy, Director of AI at Tesla

29. Building AI apps

34. Training vs Inference

35. Training vs Inference
Learning vs Answering

36. Inference
Getting a new data sample to infer an answer

38. Inference
Runs faster than Training
Models process one input at a time

39. Inference matches the FaaS model
Enough code for a function
Each function processes one request at a time
function (input) {
//1. download and cache model
//2. return inference(input)
}

40. Additional FaaS benefits
It's FaaSter to deploy the code directly
Never pay for idle
Low maintenance overhead

41. Real apps integrate multiple
algorithms into workflows

42. Real apps integrate multiple
algorithms into workflows
reusing existing functions

43. Demo
• Jupyter Notebook
https://github.com/akh64bit/qconsf
• AI Composition
http://opensource.adobe.com/adobe-sensei-ai-functions/

45. Cast
• FaaS Platform - Apache OpenWhisk
openwhisk.org
• Workflow - Apache OpenWhisk Composer
github.com/ibm-functions/composer
• Editing AI Action - JupyterLab Notebook
jupyter.org
• Deploying AI Action - JupyterLab Notebook

49. Software 2.0
Model + Code

50. Software 2.0
JupyterLab - assist in model development
Functions - assist in deploying the model

51. Software 2.0
ML Engineers collaborate with software
developers

52. Software 2.0
With FaaS it's easy to deploy a new
AI Model-as-a-Function

53. Conclusions
FaaS platforms are still maturing

54. Conclusions
FaaS platforms are still maturing
It's FaaSter to deploy AI models

55. Conclusions
FaaS platforms are still maturing
It's FaaSter to deploy AI models
Build more services, pay less