0:11

Hello, and welcome to the Data Engineering

0:13

Podcast, the show about modern data management. Introducing

0:17

Routerstack Profiles. Routerstack

0:19

Profiles takes the SAS guesswork and SQL

0:21

grunt work out of building complete customer

0:24

profiles so you can quickly ship actionable,

0:26

enriched data to every downstream team. You

0:29

specify the customer traits, then Profiles runs

0:32

the joins and computations for you to

0:34

create complete customer profiles. Get

0:36

all of the details and try the

0:39

new product today at dataengineeringpodcast.com/ Routerstack.

0:42

Data lakes are notoriously complex. For

0:45

data engineers who battle to build

0:47

and scale high-quality data workflows on

0:49

the data lake, Starburst powers petabyte-scale

0:51

SQL analytics fast, at a fraction

0:53

of the cost of traditional methods,

0:55

so that you can meet all

0:57

of your data needs, ranging from

0:59

AI to data applications to complete

1:01

analytics. Trusted by teams of all

1:03

sizes, including Comcast and DoorDash, Starburst

1:05

is a data lake analytics platform

1:07

that delivers the adaptability and flexibility

1:09

a lake has ecosystem promises. And

1:12

Starburst does all of this on an

1:14

open architecture, with first-class support for Apache

1:16

Iceberg, Delta Lake, and Hoodie, so you

1:19

always maintain ownership of your data. Want

1:22

to see Starburst in action? Go

1:24

to dataengineeringpodcast.com/Starburst and get

1:26

$500 in credits to

1:29

try Starburst Galaxy today, the easiest and fastest

1:31

way to get started using Trino. Your host

1:33

is Tobias Macy, and today I'm welcoming back

1:35

Tarush Agarwal to talk about what he and

1:37

his team at 5xData are building to improve

1:40

the user experience of the modern data stack.

1:42

So Tarush, can you start by introducing yourself

1:44

for folks who haven't heard any of your

1:46

past appearances? Of course.

1:48

Hey, Tobias. Good to be on the show again. I

1:51

think this is my third time. My background,

1:53

I'm the founder of 5x. My

1:56

backgrounds have been very focused in the data space, got

1:58

to be one of the first data engineers. at Salesforce,

2:00

one of the first data engineers in the world. Most

2:03

recently ran data for WeWork. We very,

2:05

very narrowly focused on

2:07

what used to be called the modern data stack

2:10

or data infrastructure. And

2:12

we sort of started 5X about two and

2:14

a half years ago, and

2:18

I'm sure we'll get all into that. And

2:20

again, for folks who haven't heard your past

2:23

appearances, do you remember how you first

2:25

got started working in data management, data

2:27

engineering, and what it is that's held your

2:29

attention this long? Yeah, absolutely.

2:31

Obviously, we've been through many,

2:33

many pivots. And when

2:36

we first started, we were actually building a

2:39

course or a program for

2:42

data, for sort of companies wanting to

2:44

invest in data. Post

2:46

WeWork, I would get a lot of

2:49

calls from CEOs looking to

2:51

bring in a head of data. And this was SMBs,

2:54

mid-sized enterprise, real estate, banks,

2:56

SaaS, e-commerce, across the entire

2:59

spectrum. And everyone was

3:01

talking about monetizing the data and

3:04

sort of data's the new oil. And

3:06

in reality, all of those conversations went like,

3:08

hey, I think you're sitting on a goldmine

3:10

of data, but the reality is, is you

3:13

have to go build this data platform, and

3:15

these are the vendors to speak to. And

3:17

it's not rocket science, but it's highly contextual.

3:19

And there's this thing called data modeling

3:22

and BI and self-service, which are really

3:24

the first steps. And

3:27

so out of that, we just

3:30

figured that many, many businesses

3:33

are trying to get started and

3:35

expertise in the space didn't quite

3:37

exist. So the first version

3:39

of 5X was really a course

3:42

to actually, this is kind

3:44

of what we've learned in the industry over the last 10

3:46

years, this is the playbook on how to go do it. And

3:49

that failed miserably. We barely

3:51

had anyone buy it. But yeah,

3:53

that was V1 of

3:56

back then, we will sort of 5X data

3:58

now just by this. In terms

4:00

of what you're building at 5X, you gave

4:02

a little bit of a backstory, how it came to be. And

4:06

the last time we spoke about your work

4:08

was in March of 2022. So we're at

4:10

about a year and a half

4:12

now, wondering what are some

4:15

of the notable changes in the

4:17

overall product focus, what you're building,

4:19

how you're pitching it, and just

4:21

the ways that the drastic shifts

4:23

in the data infrastructure ecosystem over

4:25

that time have influenced the ways

4:27

that you're thinking about that problem.

4:30

Yeah, I know. Great question. A

4:35

few years ago, pre-modern data stack,

4:37

we had everyone using

4:39

Informatica. And still today,

4:42

many, many, many enterprises are on

4:45

Informatica. And some big

4:47

advantages of Informatica was

4:49

it's an end-to-end platform. You can

4:51

do everything on it. You can

4:53

ingest and transform, and you

4:56

have all of your capabilities inside a

4:58

single platform. And that was really powerful.

5:00

Now, the users of

5:03

data teams hated the product because

5:06

it just isn't a very robust

5:08

product, very easy to use. And

5:11

on the banks of that, we had

5:13

all of these different vendors go

5:16

after individual pieces. So

5:18

we had the five strands of the snowflakes, and the

5:20

DBTs, and the BI's, and reverse

5:22

ETL, and 1520 other categories.

5:26

And fast forward a few years, and

5:28

we have 500 vendors in sort

5:30

of 30 different categories. And

5:33

today, what's happened,

5:35

especially in the last 18 months, is

5:37

that blank checks for

5:40

your data team, just because your

5:42

investors wanted you to invest in

5:44

data, are no longer a thing.

5:46

And CFOs are getting involved, and

5:48

data teams have been hit pretty

5:50

hard, and they need to show ROI. And

5:53

the analogy we use is what

5:55

happened from the one platform is

5:58

the analogy we use today is... to

8:00

their product, whether it makes sense or not. And

8:03

that also adds additional

8:05

strain and infrastructure

8:07

requirements around the data platform. And

8:09

I'm curious how you've seen that

8:12

influencing the ways that teams are

8:14

thinking about data, talking about data,

8:17

building around data, and whether the

8:19

outside view of, oh,

8:21

everybody's adding AI is realistic, or if it's

8:24

more just that maybe there are a couple

8:26

of toy projects that are maybe skunk works,

8:29

and the core of the business is still

8:31

just standard, let's get our business

8:33

and reporting and maybe a couple of ML models,

8:35

and we're focused on just this core capability

8:38

and AI is somewhere off in the wings,

8:40

maybe it'll become a thing later on. That's

8:43

a great question. And I think there are

8:45

a few different ways to look at it. I think for

8:48

sure we are seeing

8:50

data teams being pressured

8:54

to have an AI strategy.

8:57

So it's very real, it's data teams

8:59

that bring this up all the time.

9:02

We're obviously very much in

9:05

the infancy of AI inside the data

9:07

world. We have some text to

9:10

SQL. What I'm actually very, very

9:12

excited about is the conversational

9:14

BI in order to get there, things

9:16

like the data modeling layer, things like

9:18

the semantic layer become even more important

9:21

to essentially give AI context in

9:24

a business. Because things like

9:26

your sort of definition of what is MRR,

9:29

how many daily active users you have are

9:31

extremely nuanced, right? So we

9:33

think the semantic layer is really, is

9:36

sort of gonna be our best shot at sort of

9:38

getting into conversational AI. I've said

9:40

all of these things, if you just look at the

9:43

data life cycle in general, it's,

9:46

I think there are two different aspects to consider. What

9:48

is sort of data

9:50

practitioners and people in the industry sort

9:52

of really talk about, VC is what

9:54

they're investing in. And a lot of

9:56

that is like future sort

9:59

of state. of data. And

10:01

what I find really interesting is companies,

10:05

just the core data stack, the five

10:07

trans, the DBTs, the sort of snowflakes,

10:10

they're just getting into sort

10:12

of adoption at the

10:15

enterprise level. They're just now

10:17

getting, I think snowflake is a little

10:19

bit ahead, but all of the

10:21

other vendors, just the core vendors are finally

10:23

starting to be adopted

10:25

now. So there's

10:27

a big delay in what

10:29

practitioners and what's happening in

10:32

the industry versus really enterprise

10:34

adoption looks like. So I

10:36

think with that lens in mind,

10:39

from the enterprise standpoint, we

10:42

are quite far away from this.

10:44

I think from the SMB

10:46

standpoint, we're looking at something in

10:48

the next, I think end

10:50

of next year too, still quite early, but

10:52

I think 2025 is I think when sort of AI

10:55

and data will start to

10:57

get really interesting. And

11:00

in terms of the modern data stack, that

11:02

was a term that started getting thrown around,

11:04

I'd say probably in the 2020 timeframe

11:07

in particular. And that

11:10

was also around the time that

11:12

the venture investment in data infrastructure

11:14

and data startups was at an

11:16

all time high. They would throw

11:18

money at anybody that had the

11:20

word data somewhere in their pitch.

11:23

And now that has

11:25

dramatically shifted where data

11:28

infrastructure is the previous

11:30

generation of interest, there's still work

11:32

being done there, there's still successful businesses,

11:34

but it's not a situation where there

11:37

is room for everybody with an idea

11:39

to get funding and run a business.

11:41

And so we're starting to see a

11:43

cycle of consolidation and the ecosystem. And

11:47

I'm wondering how you're seeing that influence

11:49

the ways that people think about what

11:51

actually constitutes the modern data stack, what

11:53

are the capabilities that are actually necessary

11:55

and required, and which were the ones

11:58

that were perhaps frivolous or may maybe

12:00

just a feature of a larger product? Yeah,

12:03

I think having 15 different

12:06

categories is probably, it

12:08

was an overkill of

12:10

going from an end-to-end ecosystem to

12:12

a fragmented ecosystem. And we're

12:15

gonna see a lot of these categories

12:17

become features, which

12:19

can be adopted by sort of other categories. So

12:21

consolidation has to happen, right? Like the way the

12:23

world stands. We're also looking at, we're

12:26

looking at a lot of data companies today, which

12:28

will sort of struggle in

12:30

the next two or two, 18 months as our runway

12:32

dries up. So we

12:34

are gonna see consolidation. And

12:36

I think that's something which, with

12:38

sort of 5x, which has always been the sort

12:40

of consolidation of the data stack, is actually

12:42

very excited about. Because what

12:45

starts to happen when you start to

12:48

consolidate again is optionality

12:51

needs to exist. Like

12:53

one BI tool, you

12:55

know, it's very difficult for Snowflake to go acquire

12:58

one BI tool because they

13:00

then get locked into the

13:02

very specific use cases

13:04

which that BI tool is good at. And

13:07

sort of different companies have different use cases. So,

13:09

you know, I think it's very

13:11

healthy for the sort of 15 categories

13:14

to potentially come down to, you know, six,

13:16

seven, eight core categories. But

13:18

inside these categories, you're still gonna have

13:21

a bunch of dominant players, right? Even if you

13:23

just look at the data warehousing space, which is

13:25

very, very mature, you know, we

13:27

have Snowflake, we have Databricks, we have

13:29

BigQuery, you know, we still have companies

13:31

on Redshift. Sort

13:33

of DuckDB is a company

13:36

which, you know, has recently made huge headlines,

13:38

you know, to think of a

13:40

fifth warehouse right now getting funded in

13:42

a big, big way in an already

13:44

pretty sort of mature space is just

13:47

more confirmation that they're gonna be many,

13:50

many different use cases. And depending on what sort

13:52

of cloud you are and what type of use cases

13:54

you are, you're gonna have solutions which

13:56

make more sense than others. So

13:59

while consoling... helps decrease

14:01

the overall footprint even

14:04

in a world with eight different categories, you're going

14:06

to want to have optionality. 5x is the

14:08

answer to that. Especially

14:10

when you look at enterprise, it's

14:13

not data stack, it's data stacks. You

14:16

have different subsidies, acquisitions,

14:18

or even different departments

14:22

using different tools. What

14:24

you find in large enterprise

14:26

companies is not are they snowflake or

14:28

data breaks? Very often they're both. When

14:31

you look at again, how do you

14:33

take now an extremely

14:35

disjointed space? What

14:38

5x allows you to do is we

14:40

have this concept of workspaces. A

14:42

workspace can belong to a department or

14:44

a subsidy or the core data team.

14:47

You're just looking at what are the vendors which

14:49

make sense for you. I can go forecast my

14:51

cost of what does my spend on my particular

14:53

snowflake instance look like. You can have shared tools

14:55

across different workspaces. Some

15:05

might have their own. These

15:08

things at the enterprise level become

15:10

really difficult to manage. Even

15:13

digital identity access to

15:15

all of these different tools. These are

15:17

permissions. How do you do this across subsidies and

15:19

who has access to what? How do you audit

15:21

this from a central team perspective? These

15:23

are a whole suite of tools

15:27

which haven't been ever addressed

15:30

by the fragmentation of the

15:32

space. You would have to

15:34

go do this yourself and go back to building

15:36

your own car analogy. At

15:39

an SMB grade, there are many tutorials on how do

15:41

you go spin up the core three or four layers

15:44

in two or three hours. This starts

15:46

to balloon extremely quickly. For

15:49

all of these different types of use cases, even

15:52

inside a consolidated world, there's

15:54

going to be a huge need for this, especially

15:56

at the mid-market enterprise level. Otherwise

18:00

we have liability problems. Now

18:02

it's we only want to invest in the data

18:04

infrastructure and the data capabilities that we know are

18:06

going to be useful because otherwise we're going to

18:08

have money problems. And I'm

18:11

curious how you're seeing that calculus

18:13

start to influence the work

18:15

that data teams are doing, the ways that

18:17

they think about building their infrastructure, you know,

18:20

whether to say yes to

18:22

all of the data requests that are coming their

18:24

way, or maybe there is a little bit more

18:26

pushback about, well, why are you asking for this?

18:28

I'm just wondering how that is influencing

18:30

the way that the data teams are operating.

18:32

I think the two elements of that, right?

18:34

Like when you bring in sort of pinups

18:37

into the conversation, what everyone

18:39

obviously talks about is the cost of data

18:41

infrastructure, right? Like again, what is the cost

18:43

of all of these tools and they add

18:45

up pretty quickly. But it's sort of second

18:47

piece, which has been sort of spoken about

18:50

less is the people in data. And we

18:52

have a part from building a very fragmented

18:54

ecosystem. And we've also invented job

18:56

titles faster than, you know, then sort of

18:58

universities can even sort of keep up with

19:01

sort of actually training people in some of

19:03

these different professions. So there's consolidation

19:05

on the infrastructure, which we have been talking

19:07

about, which is happening. But I think we're

19:09

also entering the rise

19:11

of the sort of data generalist

19:14

and having, you know, do more

19:16

with less is a theme, which

19:18

is being sort of universally applied.

19:21

And we're seeing with consolidation,

19:23

there's more of a need

19:25

for, you know, people

19:27

working in the data realm to be

19:29

able to, you know, go manage the

19:31

platform, do ingestion, do modeling, do

19:34

BI. And, you know, we're going

19:36

back to, you know, sort of consolidation

19:38

across roles. And, you know, we're going to see

19:40

the rise of like much leaner teams. And again,

19:43

you know, in the sort of 2019,

19:45

sort of 2020, and I was guilty of

19:47

this myself for sure when I was

19:49

running data teams, you know, in some

19:51

ways, the sort of metric between sort

19:54

of data leaders was, you know, the size of

19:56

your data team and, you know, how big, you

19:58

know, you know, what teams do you have? have

20:00

and what different roles are you bringing

20:02

in, what are all these different use

20:04

cases and these people with very special

20:07

skill sets. I think that's reversing and

20:09

you're going to see the

20:11

rise of lean teams, which

20:14

are just way more efficient because

20:16

you're not paying the most

20:18

expensive tax in people, which is the communication

20:20

tax as you build larger and

20:22

larger teams and just being able to do more

20:25

because finely tooling is at a

20:27

point where it's more mature and

20:30

again with the

20:32

consolidation, just one person is just able to

20:34

do a lot, is to be way more

20:36

end to end. So I think

20:38

we're actually seeing it from both these different

20:40

perspectives in terms of there being layoffs and

20:42

I think data teams have been hit particularly

20:44

hard as well as more pressure

20:47

on bringing costs down across infrastructure as a

20:49

whole. Particularly as you start

20:51

talking about enterprise and coordination across

20:53

teams and across business units then

20:55

you start bringing in the conversations

20:57

around things like data fabric, data

20:59

mesh, these architectural principles, data as

21:01

a product. I'm wondering if you've

21:04

seen that start to come to

21:06

fruition and teams are actually building

21:08

that and they're realizing the value

21:10

that is promised by these approaches

21:12

or if it's largely been something

21:15

that is maybe interesting and not as

21:17

effective or harder to put into practice

21:19

and a lot of confusion. I'm just

21:21

wondering how these architectural ideas are also

21:23

influencing the ways that teams are thinking

21:25

about building their systems. I think at

21:27

the risk of being very controversial over

21:29

there, are these features, are these just

21:33

processes and thinking about how

21:35

to go structure your data teams

21:37

or are these actually entire categories

21:40

and just going into a few

21:42

of them, we don't have any

21:44

problem with your next catalog tool or

21:46

observatory tool or mesh or semantic layer. They

21:48

make sense as features but the idea is

21:50

introducing one more tool into the equation, one

21:53

more place where your team has to log

21:55

in, one more vendor you deal with,

21:57

one more platform where you have

21:59

to. go sort of do digital identity. The answer

22:01

of hey we want to go solve this problem and

22:03

we're going to build a new platform on top of

22:05

your existing platforms to go do it. That's

22:08

sort of not the answer and I

22:10

think a lot of these will potentially

22:12

be rolled into sort of bigger platforms

22:15

in sort of consolidation. But

22:17

again going back into the enterprise use cases

22:19

and again you know we get

22:21

to speak with some of the biggest data consultancies

22:24

in the world and what

22:26

they tell us is 80 90 percent and a lot of

22:30

these consultancies obviously do a lot of

22:32

referrals and do a lot of recenter

22:34

and sort of speaking to them with

22:36

it you know 80 90 percent of

22:39

their referrals are going into just core

22:41

four layers ingestion storage modeling reporting. So

22:43

you know sort of going back to

22:45

kind of what I said earlier you

22:47

know we some massive enterprise

22:49

companies which are implementing sort of these sort

22:51

of solutions. I think as we get into

22:53

this year and next year and the year

22:55

after that I think some of those renewal

22:58

conversations are going to be extremely difficult conversations

23:00

for some of these platforms to have

23:02

because the hype cycle of great you

23:05

have a huge mess and

23:07

this one tool is going to come and

23:10

solve everything and fortunately having a

23:12

layer on top of everything else is

23:14

not the answer to that. You shouldn't

23:16

have to throw away the database to

23:19

build with fast-changing data. You should be

23:21

able to keep the familiarity of SQL

23:23

into the proven architecture of cloud warehouses

23:25

but swap the decades-old batch computation model

23:28

for an efficient incremental engine to get

23:30

complex queries that are always up to date. With

23:32

Materialize you can. It's the only true SQL

23:35

streaming database built from the ground up to

23:37

meet the needs of modern data products. Whether

23:40

it's real-time dashboarding and analytics, personalization

23:42

and segmentation or automation and alerting,

23:44

Materialize gives you the ability to

23:46

work with fresh, correct and scalable

23:49

results all in a familiar SQL

23:51

interface. Go to dataengineeringpodcast.com/

23:54

materialize today to get two weeks free.

24:00

to maybe the individual team scope

24:02

and building the infrastructure components. I'm

24:04

curious, what are the main points

24:06

of friction or the most difficult

24:09

decisions that they have to make

24:11

as far as how to implement

24:13

their data platform? Yeah. You

24:15

know, I think what's happening now is we're seeing

24:18

companies who have a platform, you know, they have,

24:20

they, they sort of built everything sort of actually

24:22

coming to us to go consolidate, right? Being like,

24:24

this is becoming really difficult to manage. And in

24:26

reality, we just want one neck to choke. So

24:28

can you actually go take over all of these

24:30

different vendors and really go consolidate this, right? And

24:33

so we're doing these exercises across, you know, mid-market

24:35

across enterprise. What we're seeing is, you know, the

24:37

first question earlier, are you, are you using the

24:39

right vendors? Because the reality of it is this

24:41

is not full-time job, right? Like we have, you

24:43

know, our entire team is sort of, is

24:45

sort of tracking the space. It's what we do on

24:47

a daily basis and we can barely keep up with

24:50

what's happening. So to think that,

24:52

you know, a data team at an

24:54

enterprise company is going to make necessarily

24:57

the best decisions on what vendors to use. And

24:59

number two is even after what vendors to use,

25:01

are you using them in the right way? You

25:04

know, how they're being set up, like

25:06

building your entire modeling layer inside LookML,

25:09

isn't probably the best thing as you think about

25:11

now reusing a lot of that modeling layer and,

25:14

you know, pushing it into data science or you're

25:16

sort of, you're sort of using reverse ETL. Again,

25:18

as some of these tools start to overlap, right?

25:20

Like something which we see in enterprise companies all

25:22

the time is, you know, a portion of the

25:25

jobs are built inside 5-Tran inside day modeling, you

25:27

know, inside the DBT integration and some ability

25:29

traditionally on DBT, some are built inside Looker's

25:31

semantic layer creates this sort of huge mess

25:33

all over the place. So the idea of

25:35

having sort of very consistent workflows, which you

25:37

can do inside a single UI is again,

25:39

becoming, it's really simplifying. It's sort of preventing

25:41

a big part of that mess in the

25:44

first place. And we think the solution to

25:46

this is, is not so much going to

25:48

be on a layer on top, but it's

25:50

really how do you, you know, set things

25:52

up in a way, which is, you know,

25:54

again, give you an end to end experience.

25:56

So You don't have a mess in the first

25:58

place. So A lot of. The data teams

26:00

again as they spend all this money building

26:02

this platform and now they're starting to question

26:05

the are way off it you know Unfortunately

26:07

we are gonna have many of them going

26:09

back to the drawing board and big like

26:11

which one of these decisions with right decisions

26:13

and I see the in the bigger question

26:15

to coming up his are we even the

26:18

right people to make a decision on what

26:20

to make sense of self self for us

26:22

so I think again he i think is

26:24

gonna be a i think we'll the direction

26:26

which were heading instead of going back to

26:28

the basics of what is. Fluff and

26:30

what is you know all of these solutions

26:33

was axes actually promised ethic and it people

26:35

are waking up and realizing that a layer

26:37

on top it isn't the answer We have

26:40

to. Gonna go back to the drawing board

26:42

and figure out how to do description of

26:44

Lulu and now digging into what you're building

26:46

and five acts ways that you are thinking

26:49

about how to beautify that experience, what are

26:51

the appropriate tools and vendors and how to

26:53

integrate them. Wondering if you can talk through

26:55

some of the learning for you've gone through

26:58

as you're building out your current aeration. Of

27:00

the platform and some of the choices

27:02

that you've made and how you have

27:04

approached that tool and vendor selection. Yeah.

27:07

You don't have the benefit action. We

27:09

are more like the apple of to

27:11

space in I said at the Android.

27:13

you know. I. Yearn half ago we

27:15

were looking at opening up legal system such

27:17

that any vendetta gonna go into bed with

27:20

us that really what we've decided to do

27:22

is is go partner with fifteen twenty or

27:24

so or vendors be will be soon be

27:26

closely with them. We integrate with them at

27:28

a deep level were integrated into the sea

27:31

as processed in a we had a be

27:33

at. We have access to be eyes which

27:35

are not obligate the eyes and roadie provide

27:37

You know a sort of much that a

27:39

beeper experience. And. Have

27:41

been. You know when we think of

27:43

the word detective abilities right? Like what

27:45

are the capabilities which your data didn't

27:47

need? Second to beating River Ctl as

27:49

a capability witting set of catalogue is

27:51

a capability. Be. I is a capability

27:54

Soon we look at negotiations in your use

27:56

case instead of capabilities managed. Looking at it

27:58

as an individual sort of. ben. Level.

28:00

But what are the set of vendors

28:02

which worked really well together and we're

28:04

sort of doubling down on existing members

28:06

who were different opinions on you know

28:08

which ones meeting we know what really

28:10

was a we have upshot audience inside

28:12

cat decreased so in a we don't

28:14

pick one vendor across a category but

28:16

you know depending on the needs of

28:18

our customers in obese be certified have

28:20

an open source offering. We tried to

28:22

have the are you know the sort

28:24

of the sort of commercial yes of

28:26

non open source or products and you

28:28

know we think about what are the

28:30

different types of use cases and what

28:32

for the use case what would be

28:34

the best vendor and the coming up

28:36

a decent good and bad. like the

28:38

this set of vendors work really well

28:40

together and increasingly more people are being

28:42

a spiral opinion on based on my

28:44

use cases what works really well together

28:46

instead of we just hired someone who

28:48

came from x y z vendor and

28:50

vividly use explains the vendor and and

28:52

I've no idea about the layer before

28:54

this in the layer after this and

28:56

we just figured out from the perspective

28:58

of that integration and unifying. The Experience:

29:00

What are the core elements of

29:02

making sure that there is a

29:04

cohesive platform feel, duties disparate tools,

29:06

and the engineering work that you

29:09

and your team of had to

29:11

do to be able to give

29:13

that a more contiguous flow? Yeah,

29:15

so you know we had this

29:17

Constable Id you know we call

29:19

it super Id or said he

29:21

said of unified Idea where you

29:23

can go operate all of these

29:25

different vendors inside a single product.

29:27

So. In. A you have an

29:29

Id on top the A warehouse you

29:31

can go in a ghost that of

29:34

said editor Dvd jobs you can in

29:36

have ingested our so we have each

29:38

set of product becomes an app and

29:40

said i experience and said some of

29:43

them get embedded know able to work

29:45

with the vendors and like on a

29:47

set of removing parts of their product

29:49

and said embedded experience that is see

29:52

a more focused embedded experienced the some

29:54

categories like you know the idea either

29:56

Core of Innocence equal I'd Eats. Which

29:58

will go build ourselves. And the idea

30:01

is you know for using a set

30:03

of a set of five drown are

30:05

you using hypothetically an airboat? The apps

30:07

will be slightly different because did on

30:09

Mac Apple's Lapis each other said if

30:11

I found her concept of Jobs and

30:13

Ethel might have a set of different

30:15

concepts. so the individual apps available inside

30:17

Id I sort of slightly different from

30:19

each other. but you have a single

30:21

place to go do this. And why

30:23

this is really powerful is that is

30:25

that you know sort of data teams

30:27

can now go figure out what are

30:29

the. Ideal workflows said of Going back to

30:31

that example of I recently the transformations inside

30:33

Five Transcendence Id B D and than some

30:36

inside said of look the mouth you know

30:38

in sort of love and you decide for

30:40

due to the concerning hard to police this

30:42

because sort of vendors on giving your feet

30:44

of linux did you put this with that

30:46

he put the sort of feature off like

30:48

that's why would they do that the one

30:50

more adoption it when more engagement across each

30:52

of their products whereas inside Five x you

30:55

know we could see log into your vendors

30:57

but again the goal is can be is

30:59

certainly gives us. A single Id experience.

31:01

So whatever your data team decides

31:03

the path you one do do

31:05

we can make those apps available

31:07

which school are widows Workflows to

31:09

this leads to a highly consistent

31:11

experience because you know eighty percent

31:14

of where someone is a lot

31:16

on a daily basis and doing

31:18

the work sort of becomes odd

31:20

of like you have to sort

31:22

of becomes deserve gold and will

31:24

flow of how things happen. So

31:26

in a from our perspective we

31:28

decided to you know initially. You

31:30

know we were a platform to a

31:33

provision we would handle the procurement billing.

31:35

he goes all of those things we

31:37

had to sing. Their digital identity literacy

31:39

could go manager users behind utilization to

31:41

go look at you spend four cars

31:43

and in a more tools and optimizing

31:45

spend. We had the security so ordered

31:48

lord small these two to going into

31:50

a single place and prayed for your

31:52

see i as to when you're to

31:54

compliance teams. So we always had dogs

31:56

are fundamentalists. Adding a big shift for

31:58

us is really com. in sort

32:00

of focusing on this unified IDE experience, sort

32:03

of such that, you know, the data teams

32:05

on a daily basis can go do all

32:07

of this from one tool. And along with

32:09

just making their life easier, because they're not

32:12

logging into five different tools, it promotes a

32:14

lot of hygiene in terms of best practices,

32:16

which can be more standardized. And, you know,

32:18

you can have more guard trails up as

32:21

to this is how we want to do

32:23

things, as opposed to a free for all

32:25

any vendor just go log in and do

32:27

it in that way. And then from that

32:30

unified ID perspective, there are a

32:32

couple of interesting elements to that. One

32:34

is engineers are very opinionated about the

32:36

tools that they want to use for

32:38

doing the work that they do. And

32:41

I'm wondering what are some of the

32:43

ways that you make that ID experience

32:45

customizable so that they can feel at

32:47

home doing the work in that context.

32:50

And also, maybe some of the

32:53

ways that you're thinking about how

32:55

do we extend that experience into

32:57

the tools that people are already

32:59

using? Yeah, that's a great question.

33:01

I think, you know, what we're going to

33:03

see is a lot more apps on sort

33:05

of five x ID, right? And apps are

33:08

experiences on how you want to basically

33:10

cooperate them. So, you know, we are

33:12

again going and building, you know, some

33:14

of the core experiences again, you know,

33:16

we see that in larger businesses, again,

33:18

there's a little bit less, you know,

33:21

inside enterprise, again, you know, I think,

33:23

again, at the SMB stage, you know,

33:25

there's a lot of flexibility, people are

33:27

free to kind of use their own

33:29

tools. And, you know, again, some, it's just

33:31

they're way faster to move and adapt. And, you

33:33

know, we see sort of circle and we see

33:35

hex and, you know, different people are kind of

33:38

using different things. And that's kind of all fine,

33:40

right? It's all very manageable, you can speak to

33:42

five different people, and you know, it again, at

33:44

enterprise, it's become sort of very different, right? Like,

33:46

so there is a certain level of flexibility and

33:48

sort of customization. And, you know, there's a lot

33:50

more stuff that we've planned in terms of the

33:53

ID experience to, you know, go make it sort

33:55

of more flexible. But I think, you know, what

33:57

we're really focusing on now is what are some

33:59

of the these core sort of use cases,

34:01

which teams are really focused on and how

34:03

do we provide a really solid sort of

34:05

unified way to go do this. So that's

34:07

really what the focus is now, but we're

34:09

going to see a lot more, you know,

34:11

app, a lot more experiences to

34:14

go sort of modify this in a way

34:16

which makes sense for you while still having

34:18

some guardrails, which the company wants to have

34:20

in order to have a consistent experience. And

34:23

from that consistent experience perspective to

34:26

what is the ideal flow that

34:30

users will experience when they say, I

34:33

have this data problem I need to

34:35

resolve either I need to onboard this

34:37

data, or I need to build this

34:39

report, or I need to ensure that

34:42

these transformations are running. I'm curious, what

34:44

are the different stages of that development

34:46

flow? Some of the ways that you're

34:48

thinking about how to manage versioning and

34:51

change management, the auditability streams that you're

34:53

integrating into that experience for managers or

34:55

administrators, and just some of the key

34:57

touch points in that overall experience and

35:00

the ways that you're thinking about building

35:02

this into a cohesive product. And you

35:04

know, it's a great question. I think

35:07

we don't want to go create the

35:09

underlying sort of tooling, right? Like the

35:12

reason we go partner with everyone in

35:14

the space is we think they are

35:17

very robust solutions out there, which do a sort

35:19

of phenomenal job in sort of what they're doing.

35:21

So we don't necessarily want to go reinvent the

35:23

wheel, right? So a lot of the versioning, a

35:26

lot of the branches, we think dbt does a

35:28

great job in

35:30

a lot of these different layers. And

35:32

again, you know, ID supports dbt natively,

35:34

we have our own version of dbt

35:36

core, which we've deployed more for

35:38

smaller customers, but we continue to

35:41

be partners with dbt at the

35:43

enterprise level. And you know, we

35:45

integrate into, we integrate into that.

35:47

So the whole premise is you

35:49

get to have the underlying component

35:52

are sort of powered by a

35:54

solution, which makes sense to your

35:56

business. And in the future, they

35:58

could be coalesced. They could be sort of. unified

38:00

experiences that you want to give people

38:02

this nice, easy flow. But you're also

38:04

working with companies who probably have already

38:06

made investments into their data platform that

38:09

maybe they don't want to get rid

38:11

of, or maybe there's going to be

38:13

a long deprecation path. I'm curious what

38:15

that overall integration and migration process looks

38:17

like where they say, we've already built

38:19

a bunch of stuff, but we also

38:22

want to be able to have this

38:24

unified experience. How do you

38:26

help to bridge that gap? Yeah, that's a great question.

38:29

This 5x platform doesn't care if you

38:31

already have a vendor or you want

38:33

us to go manage your

38:35

vendor relationship and go buy that vendor from

38:37

us. You can either buy vendors from us

38:39

in a simple five billing, or

38:41

you can go import your own vendors and

38:44

have you on vendor relationships. It

38:46

doesn't make a difference. We have a bunch

38:48

of hybrid where they might have a billing

38:50

relationship with a few vendors and reintroduce a

38:52

few others, and that works great too. What

38:54

we're seeing more is we are very deliberate

38:56

on the partners we choose to work with.

38:59

We want extremely adapted partners

39:01

on our ecosystem because again, we realize

39:03

we're not going to go integrate with

39:05

500 different vendors. We're going to have

39:07

some vendors which we think are extremely

39:09

widely adopted, that these vendors are going

39:12

to stand the test of time and

39:14

we bring them on. What

39:16

we see is that it's rare

39:18

unless they're not using

39:20

a warehouse and they're doing stuff on S3

39:22

and they have Spark jobs on top of

39:24

it. It's a completely different paradigm shift. In

39:26

general, people on the warehouse world, and we

39:29

support all four of the big warehouses,

39:31

we support Snowflake BigQuery Redshift

39:33

and next year integrating with Databricks

39:35

is a big focus of us. We

39:38

support all of the big players out of

39:40

the box. We support all of

39:42

the big players out of the box. Very

39:45

often when companies want to move to

39:47

us, it's not that they are that...

39:50

For the most part, we're able to support most

39:52

of the vendors which they are already working with.

39:55

That's one piece. But again, I think we're seeing

39:58

more and what we're going to see a lot. more

40:00

next year is sort of companies coming to

40:02

us and be like, this is our data

40:04

platform. Can you actually go consolidate all of

40:06

this, even the sort of, even the sort

40:09

of vendor relationships? And, you know, we want

40:11

to have one neck to choke. So you

40:13

handle all of that. And we can spend

40:15

100% of our time focused on actually delivering

40:17

data without having to manage the vendor relationships.

40:20

In terms of those vendor relationships,

40:22

and the onboarding work, the integration

40:24

work, you mentioned that you decided

40:27

fairly early on that it wasn't

40:29

just going to be an open

40:31

ecosystem. Anybody can come in and

40:33

be part of this experience. What

40:36

are your criteria for deciding which tools,

40:38

which vendors will be incorporated into that

40:41

platform, the work that you have to

40:43

do to be able to integrate

40:46

and expose that vendor and hook them

40:48

into the overall experience.

40:50

And I'm particularly interested in how that

40:53

factors in for those vendors that don't

40:55

fit cleanly into one category or another.

40:57

And there's overlap between them. Yeah, it's

40:59

a great question. We evaluate vendors in

41:01

like four different criteria. Obviously, number one

41:04

is the technology, you know, what they're

41:06

solving, what their product is. Number two

41:08

is their roadmap. So where are they

41:10

going in the future? What's becoming more

41:12

and more relevant for them? Number three

41:15

is the

41:18

sort of partnership. Do we have

41:20

alignment inside? Do we have a

41:23

deep partnership? Do we have the sales

41:25

level? Do we have other sales customer

41:27

support at the product

41:29

level? And number four is just there's some

41:31

sort of categories which we haven't entered, right?

41:33

We think they still might be relevant, but

41:35

we don't have a really good opinion on

41:37

one way that category is going and, you

41:40

know, how do we look at it? Sort

41:42

of fourth one is sort of how do they fit

41:44

in into the sort of general macroeconomic sort of climate.

41:47

We use this sort of criteria. Again, we

41:49

want to partner with some vendors which we think

41:51

are on the uprise. And we want to partner

41:53

and we want to build sort of deep

41:56

integrations into these vendors. And

41:59

so we can go. provide the best experience

42:02

for our customers on 5x. I

42:05

think for vendors

42:08

who are across multiple different

42:10

categories, we are starting

42:12

to see that. For example, we

42:15

are starting to partner

42:17

heavily with Ruddestack and

42:19

they do the CD pieces,

42:23

but they also have a reverse ETL offering.

42:26

I think all

42:29

of this is on the case-to-case

42:31

basis. We are partnering with an

42:33

end-to-end platform called Peak.ai, which is

42:35

an end-to-end experience across data science.

42:38

We mainly focus on the data

42:40

engineering and analyst

42:42

personas. Our partnership

42:44

with Peak is more

42:47

so that a lot of the customers who

42:49

want a new data science need a data

42:52

engineering persona and a lot of our customers

42:54

who have infrastructure from 5x

42:56

ask about a data science platform. It's

42:59

a little bit disjoint, but

43:01

again, we already started to

43:03

see multiple different categories overlap.

43:05

For example, workflow manager is

43:07

something which is consistent across both. Both

43:10

data science use cases as well as

43:12

data entering use cases need

43:15

workflow manager. As we

43:18

really now get into

43:20

going deeper into all of this, it's

43:23

not going to be as clean.

43:25

It never is. It's going to be

43:28

overlap. I think our product and engineering

43:30

teams are starting to think

43:32

about these things. In your

43:35

work of building the 5x data platform,

43:37

working with your customers, working with vendors,

43:39

what are some of the most interesting

43:41

or innovative or unexpected ways that you've

43:43

seen your product used? What

43:46

we very intentionally decided to do a

43:48

few months ago is double down on

43:50

our consultancy. We've always had

43:52

a small consultancy to sort of go

43:55

help customers. Again, as

43:57

we look at the enterprise

43:59

landscape, there's a lot of

44:01

fragmentation there too because a particular

44:03

vendor is going to go sell you their

44:06

product and when you ask for help on it they're going

44:08

to introduce you to an SI integrator

44:10

or a sort of consultancy. And

44:13

if we really want to make sure that people

44:16

are doing things correctly, you know, being

44:18

able to actually offer services and, you know,

44:20

help out with some of these implementations or

44:22

as needed to sort of bring in the

44:24

expertise, we just think is just part

44:28

of helping our customers go get value

44:30

from data. So we're partnering

44:32

with a bunch of our

44:35

vendors, again, a deep partnership where

44:37

we're also becoming

44:39

SI providers and can go do

44:41

the implementations. And

44:44

I think so we have a

44:46

subset of sort of customers where we're

44:48

doing end-to-end data as a service. We

44:52

sort of give you the platform and they're using

44:54

us to go build this

44:56

sort of reporting layer on top of it. And

44:59

we recently did something with this sort

45:01

of restaurant chain, you know, they have

45:03

50 to

45:06

100 different sort of locations. They're

45:08

inside the QSR category. They

45:12

for lack of expertise and, you know,

45:14

not having sort of data people in

45:16

there, they were used to, you

45:19

know, the analytics they got from Uber

45:21

Eats and restaurant 365 and, you

45:24

know, sort of Postmates and, you know, all

45:26

these different things. And, you know, their entire

45:28

marketing agencies, which just go focus on that

45:30

category and they do what they

45:32

do. And, you know, obviously it's a big business.

45:35

And what we were able to do with

45:37

them in just a few months is the sort of level

45:40

of sort of data and, you know, the

45:42

insights we could get in the analytics perspective.

45:45

It was something which they had

45:47

sort of never seen before. So much so

45:49

that sort of marketing agency, which,

45:52

you know, works with sort of 1,700 restaurants

45:56

was, you know, just completely sort of shocked

45:58

and blown away by. They now

46:00

want to sort of go do this across all

46:02

of these sort of different verticals. I

46:04

think we're getting exposed to, you know,

46:06

sort of use cases and sort of company types,

46:09

which are really interesting was, you know, sort of

46:11

very large, but again, haven't

46:13

had the

46:15

appetite or the

46:18

conviction or the expertise to actually

46:20

go make those investments in data.

46:23

Because as a data industry, we haven't made

46:25

it easy to go make those investments in

46:27

data. Sort of given

46:30

the fragmentation, but even just given like, you

46:32

know, the support and implementation needed to actually

46:34

go get value from your product. So

46:37

we're seeing a lot more of these use cases,

46:40

which we don't wear industries

46:42

which previously haven't entered the

46:44

ecosystem are now able to

46:47

be completely disrupted, because

46:49

what we're able to do is something

46:51

which is, for

46:53

lack of better words, no one else is able to do.

46:57

And another interesting aspect of what

46:59

you're building is that because you

47:01

have this unified experience, the incremental

47:03

cost of adoption for new tools

47:05

or new capabilities is much lower

47:07

for teams than it would be

47:09

if they had to go out

47:11

and do that evaluation process, do

47:13

the integration process themselves where it

47:15

turns from, oh, I

47:17

want to use dbt or some other

47:20

tool or vendor. Now I

47:22

just say, click a button, log in, start working with

47:24

it versus Oh, now I have to spend six months

47:26

going through that whole process. And then

47:29

maybe if you are in a larger

47:31

organization, you also have to do some

47:33

selection of source paperwork, get the funding,

47:35

etc, etc. And I'm wondering how that

47:37

influences the ways that teams are approaching

47:40

that process of saying, Oh, I want to add

47:42

this new capability. I want to start doing this

47:44

new thing. Yeah, that's a good

47:47

question. I want

47:49

to answer it in once you

47:51

get used to buying a car, it

47:53

becomes very difficult to go by car parts ever

47:55

again. So once

47:57

sort of customers see how. easy

48:00

it is to go on board a vendor and have

48:03

our expertise on making some of these

48:05

decisions and helping out with implementation, they

48:09

have a much lower barrier

48:12

to, hey, what about the next tool?

48:14

Because it's not as daunting of a process

48:16

every single time. So we

48:19

have sort of companies which sort

48:21

of start small and they add capabilities as they

48:23

need them with time and they're able to, I

48:25

like to use the word, do

48:28

it in a hilariously more efficient manner

48:31

as opposed to the

48:33

way companies do this today. And

48:36

again, for us what's very important is being

48:38

able to look at this very holistically,

48:40

not just from the sort of vendor

48:42

standpoint and the decision making standpoint, but

48:44

also the implementation and making sure they're

48:46

getting value from it. How do we

48:49

provide customers with the most amazing experience

48:51

to basically go do

48:53

this? And we're playing

48:55

a long game over here, right?

48:58

We want to make sure customers are getting value from data

49:00

because the way we look at

49:02

the world is we can go sell all of these

49:04

different tools and you can go do

49:06

that, but if customers are not getting value from data, at

49:09

some point it's going to

49:11

come back, right? Like either those tools are going to get

49:13

asked, either the data team is going to get asked. And

49:17

we're not looking to go make a short term buck

49:19

by sort of going and sort

49:21

of adding a new tool over there. We

49:23

want to make sure that we're looking

49:25

at this very sort of holistically and we're

49:28

just playing a sort of longer game and we

49:30

want to create the best experience for our customers

49:32

and data teams. And

49:34

in your experience of building this product,

49:37

building the company, working with customers, what

49:39

are some of the most interesting or

49:41

unexpected or challenging lessons that you've learned

49:44

in the process? You

49:47

know, people sort of

49:49

two years ago called us crazy because

49:51

we want building a new

49:53

category. We want building what

49:56

VCs call a product, which is

49:58

this is a category. And, you know,

50:00

this is where you go sell and you know, you have,

50:03

you know, you sell your product and this

50:05

is your ACV and we are something

50:07

completely different to how the entire industry

50:09

works. And everyone

50:11

called this idea of going and

50:13

consolidating the stack in like by building a layer on

50:16

top of it, all sorts of names and all sorts

50:18

of, all sorts of craziness. And I

50:21

think the last few months

50:23

have been truly exciting because a lot

50:26

of people are really seeing the value

50:28

in this now. So, yeah, I

50:30

mean, it was, it was sort of,

50:32

it was quite challenging to go get our first

50:34

integrations. It was sort of challenging to, you know,

50:37

explain data teams when you had unlimited

50:39

funding, why they don't want

50:41

to go and build

50:43

their own platform and manage this forever.

50:45

And, you know,

50:48

it's been really reassuring

50:50

and I don't even think

50:52

that's the right word, but it's been, it's

50:54

just very, just sort of very

50:56

grateful for people to see the

50:58

value in what we're doing now.

51:02

So, yeah, just makes a lot of the

51:05

sort of decisions we made to stay on

51:07

our path back then. Just,

51:10

you know, we see a new wave of people really excited to

51:12

show off to work every day. And

51:14

for people who are in the process

51:17

of tool evaluation, maybe they're doing their

51:19

own integration work across different vendors in

51:21

the modern data ecosystem, what are the

51:24

cases where 5x is the wrong choice?

51:27

Great question. 5x is not for you if you

51:30

are not using a warehouse or don't plan

51:32

to use a data warehouse. We

51:34

are built on a data warehouse first approach. Apart

51:37

from that, I think we're relevant

51:39

across the entire lifecycle of

51:42

your data team. And what I mean by that is for

51:45

SMB companies, we have a program

51:47

where we wave the cost of

51:49

the 5x platform, you

51:51

can still take advantage of provisioning

51:53

vendors from us at

51:55

less price or even cheaper and then

51:58

using 5x to go operate all your vendors. does so

52:00

in some ways it's a no-brainer for building

52:02

a platform from scratch today to build it

52:04

on 5x, for mid-market

52:07

and enterprise companies, other

52:10

advanced tools in our suite as

52:14

we spend discussing on this podcast becoming

52:16

extremely relevant. So

52:18

you know again a no-brainer to use us

52:20

to make your data teams more efficient. Overall

52:23

I think if you are using a warehouse

52:26

first approach using 5x

52:29

to really simplify the management and operation

52:31

of your platform is

52:33

something worth doing. And

52:36

as you continue to build and iterate on

52:38

your product and business what do you have

52:40

planned for the near to medium term or

52:43

any particular projects or problem areas you're excited

52:45

to explore? I think

52:48

you know we're just

52:50

focused on the basics right like sort of getting our

52:52

platform experience to be incredibly

52:55

powerful you know as we

52:57

get more daily active users to go use to

53:00

go operate inside our platform you know big

53:02

focus for us you know

53:04

sort of enterprise and you know enterprise

53:06

readiness you know sort of everything which

53:08

happens everything you need to

53:11

basically go after that segment. We

53:13

are not looking you know apart from

53:15

Databricks next year we're

53:17

not really looking to expand our

53:19

vendor footprint drastically. We want

53:22

to double down on our existing partnerships and

53:24

you know build the best experience all the

53:26

way from sort of SI work

53:28

to you know integration into you know

53:31

they sales teams into integration into their

53:33

product and engineering and at

53:36

a macro level we're super focused

53:38

on sort of profitability as

53:40

a business so you know we're playing a long

53:42

game and I came

53:45

from WeWork I came from a company which raised a

53:47

lot of money and you know some of my lessons

53:50

over there is you want

53:52

to go figure out what business you have

53:54

sooner rather than later so we want to

53:57

be around for a long time we're playing a long game. And

54:00

we want to get the business in a position which we can

54:02

go to that. Well

54:04

for anybody who wants to get in touch with

54:06

you, follow along the work that you and your

54:08

team are doing. I'll have you add your preferred

54:10

contact information to the show notes. And as the

54:13

final question, I'd like to get your perspective on

54:15

what you see as being the biggest gap in

54:17

the tooling or technology that's available for data management

54:19

today. I

54:21

think again, you know, we spoke about a lot

54:23

of those things about if you

54:25

look at AWS, AWS is a collection

54:27

of 50 different services, but they give

54:29

you a really cohesive experience to go

54:31

manage it, right? A single place to

54:34

provision, single digital identity, billing, migration, they

54:36

give you cost optimization and

54:39

they give you role-based access control. All

54:41

of these different things have just

54:43

been hilariously missing inside the data

54:45

ecosystem. So again, the glue which

54:47

actually connects all of this together, and I still

54:49

continue to think that that's one of

54:52

the biggest missing pieces in the space. Well,

54:54

thank you very much for taking the time

54:56

today to join me and share the work

54:58

that you're doing at 5xData. It's definitely a

55:01

very interesting problem area that you're

55:03

trying to address, interesting product that you're

55:05

building around it. So appreciate all the

55:07

time and energy that you and your

55:09

team are putting into making the modern

55:11

data ecosystem a more attractable and approachable

55:13

problem space. So thank you for the

55:15

work you're doing there for your time

55:17

and I hope you enjoy the rest

55:19

of your day. Thank you very much for having me.

55:22

Thank you for listening. Don't forget to check out our other shows, Podcast.init, which

55:25

covers the Python language, its

55:28

community, and the innovative ways

55:30

it is being used, and the Machine Learning

55:32

Podcast, which helps you go from idea to production of machine

55:34

learning. Visit

55:36

the site at dataengineeringpodcast.com to subscribe to the show,

55:39

sign up for the mailing

55:49

list, and read the show notes. And

55:51

if you've learned something or tried out a product from the show,

55:53

then tell us about it. Email

55:56

host at dataengineeringpodcast.com with

55:58

your stay. other

56:00

people who are in the show, please leave a review on

56:02

Apple Podcast.