The Fifth Elephant 2017

On data engineering and application of ML in diverse domains

How to engineer a personalization system that can handle Paytm scale

Submitted by Harinder Takhar (@harindertakhar) (proposing) on Tuesday, 4 April 2017

This is a proposal requesting for someone to speak on this topic. If you’d like to speak, leave a comment.

videocam_off

Technical level

Advanced

Section

Full talk for data engineering track

Status

Submitted

Vote on this proposal

Login to vote

Total votes:  +11

Abstract

When we say we value customer experience we meant it! When you have to personalize every pixel on the app, your standard caching techniques go out of window and you need very fast and scalable system that can generate content for users in unnoticeable time. In this talk we will share how did we build our real time personalization engine which evaluates and serves over 10 billion recommended products for different channels reliably. We will share how did we use akka clustering for building this , our learnings from use of elastic cache and more importantly how did this impact our design decisions for the next service we built.

Outline

When we say we value customer experience we meant it! When you have to personalize every pixel on the app, your standard caching techniques go out of window and you need very fast and scalable system that can generate content for users in unnoticeable time. In this talk we will share how did we build our real time personalization engine which evaluates and serves over 10 billion recommended products for different channels reliably. We will share how did we use akka clustering for building this , our learnings from use of elastic cache and more importantly how did this impact our design decisions for the next service we built.

Speaker bio

We will be deciding this once the talk is accepted.

Comments

  • 1
    Zainab Bawa (@zainabbawa) Reviewer a year ago (edited a year ago)

    The draft slides need to explain why Paytm’s real-time personalization engine is interesting for audiences who do not have use cases similar to Paytm’s. What is the big picture that the audience can take away from this presentation?

  • 1
    Govind Kanshi (@govindsk) a year ago

    This would be wonderful talk to share how personalization pipeline requires different tiers of data and how their recency of occurence (specially in case of payment) matters or affects the platform.
    More than akka clustering there must be Model buildout, validation, “freshness or variety of suggestions” to ensure better chances of “hooking in”. That would be of better interest in terms of the challenges of how to measure the success and tweaking for various channels and users combination. Do you plan to share “non proprietary part - something which people can carry away for such large scale customer base.

Login with Twitter or Google to leave a comment