Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Build a serverless recommendation engine in 72 hours

Build a serverless recommendation engine in 72 hours

Samuel James

October 16, 2019
Tweet

More Decks by Samuel James

Other Decks in Technology

Transcript

  1. Build a serverless recommendation engine in 72 hours
    @samueljabiodun | Samuel James

    View Slide

  2. @samueljabiodun
    Who Am I?
    − AWS Solutions Architect (Associate)
    − Software Engineer @ Architrave Gmbh
    ✦ PropTech company of the year 2018
    ✦ Over €12 billion in annual transaction volume (including Germany's largest
    single transactions: Sony Centre 2017 & Frankfurt Omni Tower 2018)
    ✦ Over 3,900 managed assets worth €80 billion

    − Technical Author @ HitSubscribe

    View Slide

  3. @samueljabiodun
    “At some point we’re going to see the first billion-dollar
    startup with a single employee, the founder, and that
    engineer will be using serverless."
    James Governor
    Co-founder at RedMonk

    View Slide

  4. @samueljabiodun
    In the beginning …

    View Slide

  5. @samueljabiodun
    Physical Machine
    ❏ Install and configure tons of packages to get your
    application up and run
    ❏ Very Expensive
    ❏ Scaling was an hassle
    ❏ Maintenance was still a problem

    View Slide

  6. @samueljabiodun

    View Slide

  7. @samueljabiodun
    Code
    Runtime
    OS
    Hardware
    Physical Servers

    View Slide

  8. @samueljabiodun
    Code
    Runtime
    OS
    Hardware
    Physical Servers
    Code
    Runtime
    OS
    Hardware
    Virtual Machines
    Code
    Runtime
    OS
    Hardware
    Code
    Runtime
    OS
    Hardware
    Serverless
    Containers

    View Slide

  9. @samueljabiodun
    Code
    Runtime
    OS
    Hardware
    Serverless
    Serverless

    …computing model that allows you to build
    and run code at scale without thinking about
    servers or infrastructures

    View Slide

  10. @samueljabiodun
    Serverless and PaaS

    What is the difference?

    View Slide

  11. @samueljabiodun
    − Scaling
    − Composition
    − Idle State
    Why PaaS Is Not Serverless (FaaS)

    View Slide

  12. @samueljabiodun
    − Scaling
    − Composition
    − Idle State
    Why PaaS Is Not Serverless (FaaS)

    View Slide

  13. @samueljabiodun
    Business Logic

    View Slide

  14. @samueljabiodun
    − Scaling
    − Composition
    − Idle State
    Why PaaS Is Not Serverless (FaaS)

    View Slide

  15. @samueljabiodun
    Why going serverless?
    − No server provisioning
    − Pay for usage only
    − Infinite scale
    − No server management
    − Quicker time to market
    − Monitoring and logging out of the box

    View Slide

  16. @samueljabiodun
    "There is no silver bullet. There are
    always options and the options have
    consequences”

    - Ben Horowitz

    View Slide

  17. @samueljabiodun
    Use cases
    − Data processing (Images,
    Videos etc)
    − Scalable static websites
    − Mobile backends
    − Real-time analysis
    − IoTs
    − Event streaming applications
    − Multi-lingual applications
    − And more…

    View Slide

  18. @samueljabiodun
    Serverless Recommendation Engine

    View Slide

  19. @samueljabiodun

    View Slide

  20. @samueljabiodun

    View Slide

  21. @samueljabiodun
    Recommendation is about suggesting relevant
    products to your users based on the knowledge
    you have users, products, and users' interactions
    with your products

    View Slide

  22. @samueljabiodun
    Recommendation is all about data

    View Slide

  23. @samueljabiodun

    View Slide

  24. @samueljabiodun

    According to McKinsey & Company, 35%
    of Amazon.com’s revenue is generated by
    its recommendation engine.

    View Slide

  25. @samueljabiodun
    "Netflix saves up to $1 billion a year via
    its personalised recommendations” 


    – Business Insider

    View Slide

  26. @samueljabiodun
    Personalised recommendations drive
    sales

    View Slide

  27. @samueljabiodun
    Data
    Collection
    Data Storage Data Analysis Data Filtering
    Recommendation Phases

    View Slide

  28. @samueljabiodun
    Data Filtering Techniques
    − Collaborative Filtering
    − Content-Based Filtering
    − Hybrid Filtering

    View Slide

  29. @samueljabiodun
    Collaborative Filtering
    Interactions of users with products (like movies
    watched, products viewed, products bought etc.

    View Slide

  30. @samueljabiodun
    Collaborative Filtering Technique in a
    Retail Site

    View Slide

  31. @samueljabiodun

    View Slide

  32. @samueljabiodun

    View Slide

  33. @samueljabiodun
    Content-based Filtering
    Focuses on properties of items. Similarity of items
    is determined by measuring the similarity in their
    properties.

    View Slide

  34. @samueljabiodun
    Content-based Filtering Technique in a
    Retail Site

    View Slide

  35. @samueljabiodun
    AWS Rekognition At A Glance

    View Slide

  36. @samueljabiodun

    View Slide

  37. @samueljabiodun

    Analysing existing (millions,
    billions of) objects on S3 using
    AWS Batch

    View Slide

  38. @samueljabiodun

    View Slide

  39. @samueljabiodun
    Hybrid Filtering
    Combines collaborative filtering and content-
    based filtering

    View Slide

  40. @samueljabiodun
    Learn about
    your users
    Gain insights in
    to your products

    View Slide

  41. @samueljabiodun
    Users do search for products
    too, how about adding a visual
    search?

    View Slide

  42. @samueljabiodun

    View Slide

  43. @samueljabiodun
    DEMO

    View Slide

  44. @samueljabiodun

    View Slide

  45. @samueljabiodun
    Summary
    − Serverless reduces your development time
    − With Serverless, you do not need to be an AI expert to have AI capabilities in your
    app

    View Slide

  46. @samueljabiodun

    View Slide

  47. @samueljabiodun
    Thanks!

    Questions?

    View Slide