Security Cloud Privacy Tech
Swami Sivasubramanian's Keynote at AWS re:Invent 2021

Swami Sivasubramanian's Keynote at AWS re:Invent 2021

AWS re:Invent is always THE event in cloud. Swami Sivasubramanian took the stage to deliver the machine learning focused keynote on day three.

This πŸ‘‡ is the Twitter thread of my coverage of the keynote…

Tweet 1/141 πŸ‘‡ Next tweet

. @SwamiSivasubram up now for the #ml keynote at @awscloud #reinvent


Tweet 2/141 πŸ‘‡ Next tweet πŸ‘† Start

Swami is setting the scene; data is everywhere. there’s mountains of it. that makes it hard to get value from it


Tweet 3/141 πŸ‘‡ Next tweet πŸ‘† Start

β€œSurvival of the most informed”, @SwamiSivasubram


Tweet 4/141 πŸ‘‡ Next tweet πŸ‘† Start

quick little nod to swimming Australia. they are using @awscloud #ml services to pull insights from their training data


Tweet 5/141 πŸ‘‡ Next tweet πŸ‘† Start

name drops for @NasdaqTech, @Philips, and @carrier … all previous customer references


Tweet 6/141 πŸ‘‡ Next tweet πŸ‘† Start

now walking through an e-commerce application example


Tweet 7/141 πŸ‘‡ Next tweet πŸ‘† Start

step 1: shove things into an RDBMS


Tweet 8/141 πŸ‘‡ Next tweet πŸ‘† Start

step 2: realize that breaks at some point. time to use specific data stores that align to your data & it’s use


Tweet 9/141 πŸ‘‡ Next tweet πŸ‘† Start

step 3: snag the streaming data too


Tweet 10/141 πŸ‘‡ Next tweet πŸ‘† Start

step 4: start asking questions to draw out insights


Tweet 11/141 πŸ‘‡ Next tweet πŸ‘† Start

step 5: realize a data lake is going to help unite all of these data sources and questions


Tweet 12/141 πŸ‘‡ Next tweet πŸ‘† Start

step 6: move beyond analysis and visualization to prediction


Tweet 13/141 πŸ‘‡ Next tweet πŸ‘† Start

…that obviously isn’t that clean a process and takes time but the gist is spot on

it’s a journey and it’s critical to use the right tool at the right time


Tweet 14/141 πŸ‘‡ Next tweet πŸ‘† Start

ooohhh, nice shout out to the need for #security and #privacy controls throughout this process


Tweet 15/141 πŸ‘‡ Next tweet πŸ‘† Start

good summary slide for the journey that @SwamiSivasubram’s example took us on


Tweet 16/141 πŸ‘‡ Next tweet πŸ‘† Start

tl:dr on the sales/marketing side: @awscloud has a ton to offer in this area


Tweet 17/141 πŸ‘‡ Next tweet πŸ‘† Start

lots of #ml adoption in the @awscloud community


Tweet 18/141 πŸ‘‡ Next tweet πŸ‘† Start

btw, I’m betting on/hoping for at least @mza appearance in this keynote!


Tweet 19/141 πŸ‘‡ Next tweet πŸ‘† Start

3 key elements to a modern end-to-end data strategy…


Tweet 20/141 πŸ‘‡ Next tweet πŸ‘† Start

1. modernize => move away from on-premises & the 90’s? 🀣 2. unify => take the data lake approach 3. innovate => create new experiences & draw new insights


Tweet 21/141 πŸ‘‡ Next tweet πŸ‘† Start

diving deeper into β€œmodernize”, we have a BMW video segment


Tweet 22/141 πŸ‘‡ Next tweet πŸ‘† Start

surprising no one, @awscloud is the platform for the BMW connected car initiative (currently 15 million connected vehicles)


Tweet 23/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram is cruising here. very quick moving keynote…


Tweet 24/141 πŸ‘‡ Next tweet πŸ‘† Start

this slide should’ve said:

Managing infrastructure, that’s so 90’s 🀣


Tweet 25/141 πŸ‘‡ Next tweet πŸ‘† Start

yup. 100%…


Tweet 26/141 πŸ‘‡ Next tweet πŸ‘† Start

little update on/pitch for @awscloud Aurora << if you’re in the market for an RDBMS solution, start here

more on Aurora at


Tweet 27/141 πŸ‘‡ Next tweet πŸ‘† Start

as usual @QuinnyPig is also covering the keynote. his thread is up at


Tweet 28/141 πŸ‘‡ Next tweet πŸ‘† Start

ohhh, new feature/service coming…you can just feel it


Tweet 29/141 πŸ‘‡ Next tweet πŸ‘† Start

yup…definitely something coming around using #ml to predict data store ops issues


Tweet 30/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE/SERVICE: Amazon DevOps Guru for RDS < it’s DevOps Guru pointed at RDS

GA today


Tweet 31/141 πŸ‘‡ Next tweet πŸ‘† Start

more on the launch at


Tweet 32/141 πŸ‘‡ Next tweet πŸ‘† Start

I ❀️ these services because they leverage the scale & insight of @awscloud to help the rest of us build better


Tweet 33/141 πŸ‘‡ Next tweet πŸ‘† Start

ohhh, are we about to get another β€œGet off proprietary databases” service?


Tweet 34/141 πŸ‘‡ Next tweet πŸ‘† Start

Amazon RDS Customer flew under the radar for me

more on the service at

<< the announcement today was that it now supports SQL Server as well. GA today


Tweet 35/141 πŸ‘‡ Next tweet πŸ‘† Start

…come on @awscloud Neptune #serverless…



Tweet 36/141 πŸ‘‡ Next tweet πŸ‘† Start

nope (or not yet?). I think we’re going in a different direction for the moment


Tweet 37/141 πŸ‘‡ Next tweet πŸ‘† Start

EXCELLENT QUESTION …and one that not enough builders are asking (especially in legacy environments)


Tweet 38/141 πŸ‘‡ Next tweet πŸ‘† Start

. @awscloud DynamoDB might be processing a few requests at any given time 🀣


Tweet 39/141 πŸ‘‡ Next tweet πŸ‘† Start

ok, getting the setup for a new feature/service again


Tweet 40/141 πŸ‘‡ Next tweet πŸ‘† Start

sidenote: this feels like more a @ajassy or @werner style of keynote. problem build => new feature/service


Tweet 41/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: Amazon DynamoDB Standard-Infrequent Access table class << this should help reduce some costs for people and allow them to keep data in DynamoDB longer

can’t wait to hear @alexbdebrie’s thoughts


Tweet 42/141 πŸ‘‡ Next tweet πŸ‘† Start

more on the new table class at


Tweet 43/141 πŸ‘‡ Next tweet πŸ‘† Start

btw, if you’re working with @awscloud DynamoDB you NEED @alexbdebrie’s book:

it’s a life saver


Tweet 44/141 πŸ‘‡ Next tweet πŸ‘† Start

also, watch one of his million+ talks on DynamoDB. only person I’ve seen to get a two-part talk at @awscloud #reinvent. that one happened last year

Tweet 45/141 πŸ‘‡ Next tweet πŸ‘† Start

back to the keynote, lots of purpose-built databases from @awscloud


Tweet 46/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram touching on the challenges around migration of databases to @awscloud now


Tweet 47/141 πŸ‘‡ Next tweet πŸ‘† Start

more on the @awscloud Data Migration Service (DMS) at


Tweet 48/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: @awscloud Database Migration Service Fleet Advisor << migration your databases en masse. it routes the data through S3 and customized migration plans to align w/the right purpose-built service


Tweet 49/141 πŸ‘‡ Next tweet πŸ‘† Start

shifting to the β€œUnify” point now…


Tweet 50/141 πŸ‘‡ Next tweet πŸ‘† Start

. @Expedia up as the customer example


Tweet 51/141 πŸ‘‡ Next tweet πŸ‘† Start

nothing in the customer example other than they are using @awscloud πŸ˜•


Tweet 52/141 πŸ‘‡ Next tweet πŸ‘† Start


Tweet 53/141 πŸ‘‡ Next tweet πŸ‘† Start

key point here is that your data should be in one place and then pulled into the right tool at the right time >> data lake


Tweet 54/141 πŸ‘‡ Next tweet πŸ‘† Start

new to the whole β€œdata lake” thing? here’s a primer from @awscloud:


Tweet 55/141 πŸ‘‡ Next tweet πŸ‘† Start

…and if you’re just getting started, use @awscloud Lake Formation and save yourself a ton of time

more on that at


Tweet 56/141 πŸ‘‡ Next tweet πŸ‘† Start

all of these strategies start with @awscloud S3. it’s the easiest, least expensive, and most performant way to store the data at scale


Tweet 57/141 πŸ‘‡ Next tweet πŸ‘† Start

# reinvent

Tweet 58/141 πŸ‘‡ Next tweet πŸ‘† Start

if you’re not using Lake Formation, your β€œlake” is probably going to end up more like a pit. that’s bad

don’t do that


Tweet 59/141 πŸ‘‡ Next tweet πŸ‘† Start

. @awscloud Athena is my 2nd favourite service (behind QuickSight) and often overlooked

more on Athena at


Tweet 60/141 πŸ‘‡ Next tweet πŸ‘† Start

ohhh, are we getting some improvements to Athena?!?!


Tweet 61/141 πŸ‘‡ Next tweet πŸ‘† Start


Tweet 62/141 πŸ‘‡ Next tweet πŸ‘† Start

. @awscloud also takes the β€œpurpose-built” approach when it comes to analytics

interactive => Athena big data => EMR ops & logs => OpenSearch real-time => Kinesis & MSK warehouse => Redshift



Tweet 63/141 πŸ‘‡ Next tweet πŸ‘† Start

πŸ‘† most of those are now available #serverless. which is fβ€”king awesome, btw


Tweet 64/141 πŸ‘‡ Next tweet πŸ‘† Start

RedShift is under utilized by builders. mainly because it’s terrifying…specially when you look at cost

the #serverless announcement yesterday will open it up to a whole new audience

more on that at


Tweet 65/141 πŸ‘‡ Next tweet πŸ‘† Start

lots of Redshift stuff. @awscloud knows that #serverless feature will intro a new audience to it, time to catch them up


Tweet 66/141 πŸ‘‡ Next tweet πŸ‘† Start

finally an #ml tie-in!


Tweet 67/141 πŸ‘‡ Next tweet πŸ‘† Start

…for those of you just joining this keynote at @awscloud #reinvent, this IS the #ml keynote

to be fair, 99% of ML is data clean up, organization, and management πŸ˜‰

Tweet 68/141 πŸ‘‡ Next tweet πŸ‘† Start

shifting the conversation to BI…QuickSight!!! :-)


Tweet 69/141 πŸ‘‡ Next tweet πŸ‘† Start

more on QuickSight Q at


Tweet 70/141 πŸ‘‡ Next tweet πŸ‘† Start

if you’re wondering why I ❀️ QuickSight so much, it’s because it opens up a lot of analytical power to a wide audience within your business

that’s amazing powerful and something we don’t consider often enough


Tweet 71/141 πŸ‘‡ Next tweet πŸ‘† Start

…also QuickSight has some very powerful sharing built in:


Tweet 72/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram’s on to @awscloud Glue while I’ve been ranting about QuickSight

my favourite AWS name, β€œAWS Glue Data Brew” << it’s true, you can use it too


Tweet 73/141 πŸ‘‡ Next tweet πŸ‘† Start

more on @awscloud Glue Data Brew:


Tweet 74/141 πŸ‘‡ Next tweet πŸ‘† Start

Neeraja up now to talk about @awscloud Redshift #serverless + QuickSight


Tweet 75/141 πŸ‘‡ Next tweet πŸ‘† Start


Tweet 76/141 πŸ‘‡ Next tweet πŸ‘† Start

walking through a use case example now…

first the informercial β€œon-premises” view 🀦🀣


Tweet 77/141 πŸ‘‡ Next tweet πŸ‘† Start

then right into the β€œdon’t worry about it” #serverless approach with the new Redshift service


Tweet 78/141 πŸ‘‡ Next tweet πŸ‘† Start

Redshift’s live data sharing capabilities lets more teams access the data without having to move it anywhere

more on the sharing features at


Tweet 79/141 πŸ‘‡ Next tweet πŸ‘† Start

QuickSight comes into play now for the data analysis at the last mile. it enables ALL teams (not just BI specialists) to query the data and draw insights


Tweet 80/141 πŸ‘‡ Next tweet πŸ‘† Start

what hits home for me with this example is the low operational overhead. almost all of the work & cost is directly driving business value

that’s the goal of the @awscloud Well-Architected Framework & these types of services make it way easier to strike that balance


Tweet 81/141 πŸ‘‡ Next tweet πŸ‘† Start

β€œYou’ve got to start treating data like an organizational asset”, @SwamiSivasubram


Tweet 82/141 πŸ‘‡ Next tweet πŸ‘† Start

we’ve moved on to culture. another customer example up, this time it’s @ADP


Tweet 83/141 πŸ‘‡ Next tweet πŸ‘† Start

lots of setup in the customer story, key point so far is that @adp handles so much data it’s like it’s a major country on it’s own


Tweet 84/141 πŸ‘‡ Next tweet πŸ‘† Start

. @adp story was about using data at scale to drive change. no real technology aspect (beyond, "we couldn't do this without @awscloud")

quick message from @slack now…


Tweet 85/141 πŸ‘‡ Next tweet πŸ‘† Start

on to innovation now...


Tweet 86/141 πŸ‘‡ Next tweet πŸ‘† Start

this slide just keeps getting bigger each year


Tweet 87/141 πŸ‘‡ Next tweet πŸ‘† Start

we’re a hop, skip, and a jump away from just a board of single pixels with just the brand’s dominant colour


Tweet 88/141 πŸ‘‡ Next tweet πŸ‘† Start

I like this stacked layer approach to explaining the @awscloud #ml service offerings

basically, you want to start at the top layer & stay there as long as possible.

when necessary, move to the SageMaker layer

…hope you never need the bottom layer πŸ€£πŸ˜‰


Tweet 89/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram’s going to start at the bottom layer


Tweet 90/141 πŸ‘‡ Next tweet πŸ‘† Start

this is all low level stuff, that’s why I recommend staying away from it

super interesting from a math/engineering perspective but also, quite a ways away from business value


Tweet 91/141 πŸ‘‡ Next tweet πŸ‘† Start

more from the low-level layer of the #ml stack on @awscloud


Tweet 92/141 πŸ‘‡ Next tweet πŸ‘† Start

more on the Trn1:


Tweet 93/141 πŸ‘‡ Next tweet πŸ‘† Start

…if you stay at the top of the stack!

(though SageMaker is amazing)


Tweet 94/141 πŸ‘‡ Next tweet πŸ‘† Start

it’s just that SageMaker required #ml specific knowledge. that’s ok if you have it. but the top layer requires near zero #ml knowledge

just hand the service something and get a result…


Tweet 95/141 πŸ‘‡ Next tweet πŸ‘† Start

ok, back to the keynote, talking about @awscloud SageMaker now. it’s an entire suite of solutions working together


Tweet 96/141 πŸ‘‡ Next tweet πŸ‘† Start

…more of a prefix than anything, SageMaker _________


Tweet 97/141 πŸ‘‡ Next tweet πŸ‘† Start

another customer story, this time it’s Aurora (not the database)

they do self-driving vehicle tech


Tweet 98/141 πŸ‘‡ Next tweet πŸ‘† Start

this Aurora:


Tweet 99/141 πŸ‘‡ Next tweet πŸ‘† Start

so far the customer story is about the promise of self-driving cars


Tweet 100/141 πŸ‘‡ Next tweet πŸ‘† Start


Tweet 101/141 πŸ‘‡ Next tweet πŸ‘† Start

this customer story is cool but I feel like it’s missing the mark. it’s not technical enough for most builders and too technical for non-builders. right in the squishy middle that leaves me wondering, β€œWhat am I supposed to take away from this?”


Tweet 102/141 πŸ‘‡ Next tweet πŸ‘† Start

I would love to get into the details of their #ml workflow or their strategy of overlaying models to deliver that self-driving vehicle

this is just an ad for the fact they are working on the tech. I wanted more


Tweet 103/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram is back up now. thankfully his energy will put this back on track. he’s still rolling strong…


Tweet 104/141 πŸ‘‡ Next tweet πŸ‘† Start

structured data, the β€œeasy” data type, is about 20% of the data #ml is working with

unstructured data is far more common

SageMaker has features to help with both


Tweet 105/141 πŸ‘‡ Next tweet πŸ‘† Start

more on Amazon SageMaker Data Wrangler:


Tweet 106/141 πŸ‘‡ Next tweet πŸ‘† Start

more on Amazon SageMaker Ground Truth at


Tweet 107/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: Amazon SageMaker Ground Truth Plus << high quality training dataset fast, and reduce data labelling cost

more at


Tweet 108/141 πŸ‘‡ Next tweet πŸ‘† Start

anything that helps with data labelling and clean up is very, very welcome


Tweet 109/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: Amazon SageMaker Studio Notebook << perform data engineering, analytics, and #ML workflows in one notebook.

GA today


Tweet 110/141 πŸ‘‡ Next tweet πŸ‘† Start

we’re cruising right along with the #ml keynote at @awscloud #reinvent

Tweet 111/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURES (under the hood):
  • Amazon SageMaker Training Compiler << 50% speed boost for training
  • …Inference Recommender << reduce time to deploy
  • … #Serverless Inference << fβ€”k infrastructure πŸ˜‰


Tweet 112/141 πŸ‘‡ Next tweet πŸ‘† Start

that’s 5(?) services or features that have made a #serverless model available to users. I ❀️ it!

should be the default for everything…


Tweet 113/141 πŸ‘‡ Next tweet πŸ‘† Start

if you didn’t see the announcement about SageMaker Canvas in @aselipsky’s keynote, you can read more at


Tweet 114/141 πŸ‘‡ Next tweet πŸ‘† Start

Allie up now to walk us through SageMaker Canvas


Tweet 115/141 πŸ‘‡ Next tweet πŸ‘† Start

the example is forecasting. something a lot of us do


Tweet 116/141 πŸ‘‡ Next tweet πŸ‘† Start

the example is a model that forecasts the need for new computers for your team (a/k/a onboarding new ppl)

3 steps:

  • access & browse the data
  • prepare the data
  • train & build the model


Tweet 117/141 πŸ‘‡ Next tweet πŸ‘† Start

so, the example is covering all computer demands, not just onboarding


Tweet 118/141 πŸ‘‡ Next tweet πŸ‘† Start

drag & drop various data sources to create one data set. SageMaker Canvas also tries to automatically clean up the data



Tweet 119/141 πŸ‘‡ Next tweet πŸ‘† Start

what, I have to click?!? ugh.


Tweet 120/141 πŸ‘‡ Next tweet πŸ‘† Start

the platform let’s you work through β€œwhat if” scenarios as well. that’s really powerful


Tweet 121/141 πŸ‘‡ Next tweet πŸ‘† Start

the real question is how broadly applicable is this beyond forecasting. how will @awscloud SageMaker Canvas handle those other use cases?

can’t wait to find out!


Tweet 122/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram back up again now...


Tweet 123/141 πŸ‘‡ Next tweet πŸ‘† Start

feels like we’re in the industrial #ml solutions now


Tweet 124/141 πŸ‘‡ Next tweet πŸ‘† Start

tweeted too soon, that was the intro to the high level view of #ml solutions


Tweet 125/141 πŸ‘‡ Next tweet πŸ‘† Start

quiet service callback, @awscloud Kendra,


Tweet 126/141 πŸ‘‡ Next tweet πŸ‘† Start

Kendra is one of several services that most builders forget even exists. that’s ok

that’s actually kind of cool. the @awscloud has grown to the point where there’s entire categories you’ll never touch as a builder


Tweet 127/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: Amazon Kendra Experience Builder << no code search application builder


Tweet 128/141 πŸ‘‡ Next tweet πŸ‘† Start

more on EMR cluster and Spark job management with SageMaker Studio:


Tweet 129/141 πŸ‘‡ Next tweet πŸ‘† Start

more on SageMaker’s new training compiler:


Tweet 130/141 πŸ‘‡ Next tweet πŸ‘† Start

more on @awscloud SageMaker inference recommender:


Tweet 131/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram’s talking about @awscloud Connect and Lex experiences


Tweet 132/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW FEATURE: Amazon Lex Automated Chatbot Designer << simplifies bot design with natural language understanding


Tweet 133/141 πŸ‘‡ Next tweet πŸ‘† Start

the Chatbot Designer sounds pretty cool. hopefully it’ll reduce user frustration


Tweet 134/141 πŸ‘‡ Next tweet πŸ‘† Start


Tweet 135/141 πŸ‘‡ Next tweet πŸ‘† Start

. @SwamiSivasubram talking about bringing more builders under the #ml umbrella. lots of work to be done in this area


Tweet 136/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW SERVICE: Amazon SageMaker Studio Lab << no cost, no setup #ml learning environment

sign up with an email address and start learning!


Tweet 137/141 πŸ‘‡ Next tweet πŸ‘† Start

πŸ‘† this is huge! it should make it significantly easier to learn #ml technologies

goes hand-in-hand with the D2L textbook:


Tweet 138/141 πŸ‘‡ Next tweet πŸ‘† Start

NEW PROGRAM: @awscloud AI & ML Scholarship Program


Tweet 139/141 πŸ‘‡ Next tweet πŸ‘† Start

Tweet 140/141 πŸ‘‡ Next tweet πŸ‘† Start

the lab is up at


Tweet 141/141 πŸ‘‡ Next tweet πŸ‘† Start

on to the @awscloud DeepRacer finale now. good luck to all!



More Content