►
Description
Today we will be interviewing EvolutionIQ and why they decided to use Dagster
Tomas Vykruta - https://www.linkedin.com/in/tvykruta/
Karan Uppal - https://www.linkedin.com/in/karanuppal/
If you enjoyed this video, check out some of my other top videos.
Top Courses To Become A Data Engineer In 2022
https://www.youtube.com/watch?v=kW8_l57w74g
What Is The Modern Data Stack - Intro To Data Infrastructure Part 1
https://www.youtube.com/watch?v=-ClWgwC0Sbw
If you would like to learn more about data engineering, then check out Googles GCP certificate
https://bit.ly/3NQVn7V
If you'd like to read up on my updates about the data field, then you can sign up for our newsletter here.
https://seattledataguy.substack.com/
Or check out my blog
https://www.theseattledataguy.com/
And if you want to support the channel, then you can become a paid member of my newsletter
https://seattledataguy.substack.com/subscribe
Tags: Data engineering projects, Data engineer project ideas, data project sources, data analytics project sources, data project portfolio
_____________________________________________________________
Subscribe: https://www.youtube.com/channel/UCmLGJ3VYBcfRaWbP6JLJcpA?sub_confirmation=1
_____________________________________________________________
About me:
I have spent my career focused on all forms of data. I have focused on developing algorithms to detect fraud, reduce patient readmission and redesign insurance provider policy to help reduce the overall cost of healthcare. I have also helped develop analytics for marketing and IT operations in order to optimize limited resources such as employees and budget. I privately consult on data science and engineering problems both solo as well as with a company called Acheron Analytics. I have experience both working hands-on with technical problems as well as helping leadership teams develop strategies to maximize their data.
*I do participate in affiliate programs, if a link has an "*" by it, then I may receive a small portion of the proceeds at no extra cost to you.
A
A
A
B
A
A
C
Sure
I'm
Tom
bakruda
I'm,
the
co-founder
CEO
of
evolution
IQ
my
background
is
technology
I've
been
doing
this
since
the
90s
I've
worked
my
way
through
a
bunch
of
different
companies,
always
working
on
kind
of
zero
to
one
Innovation
projects.
I
spent
almost
10
years
before
Evolution
IQ
at
Google,
my
focus
there
initially
was
large-scale
ROTC
back-end
services
and
then
the
second
half
of
my
tenure
was
all
about
machine
learning.
C
I
worked
on
the
autonomous
cars
I
worked
on
various
ml
projects,
including
like
fighting
fraudulent
Bank
listings
on
Indian,
Google
Maps,
a
lot
of
interesting
stuff
and
eventually
you
know
I
started
looking
for
opportunity
outside
of
digital
advertising.
You
know
I
love
the
technology
and
I
thought.
Surely
there
must
be
some
big?
You
know
Big
Data,
big
kind
of
human
human
problem
out
there
beyond
what
Google's
doing
and
that's
what
brought
us
to
Ocean
IQ.
A
A
Yeah,
exactly
I
was,
like
oh
I,
think
I
think
I
feel
like
it's
like
150
200
is
like
when
you
start
going
from
series
B
to
like
whatever
is
after
that
at
that
point
right
like
serious,
C
and
just
just
onwards,
yeah.
Well,
thanks!
So
much
for
for
jumping
on
you
know,
I
reached
out
to
you
guys
both
because
you
know
you
referenced
picking,
dagster
I,
remember
seeing
the
article
about
it
and
I
think
at
that
time,
I
was
like
doing
a
deeper
dive
into
airflow
and
I
was
like
oh
well.
A
Everyone
I
talked
to
at
that
point
was
like.
Oh,
we
looked
at
air
or
we
looked
at
dagster
and
prefect,
but
we
went
with
airflow.
So
that's
one
reason
when
you
guys
went
with
Dexter
I
was
like
okay
I'm
really
curious
in
terms
of
like
why
you
guys
picked
daxters.
That's
really
what
I
wanted
to
focus
on
kind
of
for
this
conversations
like
well
what
what
about
Dexter
really
stuck
out
but
and
I?
A
C
Yeah
for
me
personally,
so
my
entry
into
machine
learning
was
interesting.
Google
was
kind
of
pioneer
of
deep
learning.
You
know
back
in
I.
Think
2013
Alex
won
this
image
of
the
competition.
You
know
he
brought
neural
net
neural
networks
back
from
the
Ice
Age
and
Google
Google
are
interested
in
this.
You
know
they
realize
this
is
a
big
opportunity
and
they
built
this
group
inside
called
machine
learning
ninjas,
and
they
took
like
15
of
us,
put
us
with
this
program
to
teach
us.
C
What
back
then
was
like
you
know,
kind
of
Rogue
science
that
nobody
understood,
and
we
were
tasked
with
learning,
machine
learning
and
learning
specifically
how
to
apply
it
to
like
large,
applied
existing
data
problems
that
we're
using
heuristic
Solutions.
You
know
using
like
2000
technology
and
I
remember.
For
me
this
was
same
kind
of
excitement
that
I
felt
as
a
kid.
When
I
first
got
into
computers,
you
know,
then
the
magic
wore
off
and
then
in
like
2013,
it
was
back.
C
A
B
Yeah,
for
me,
it
was
about
10
years
or
so
ago.
I
was
doing
speech,
processing
work
a
lot
of
it
during
my
Master's
after
my
masters
for
a
bit
and
I
did
not
think
of
that
as
getting
into
Data
I,
just
thought
of
it
as
getting
into
doing
speech
processing,
because
that's
what
I'm
interested
in
and
from
there
I
started
doing
a
lot
more
machine
learning
Because
by
the
time,
I
started
doing
speech
processing.
It
was
basically
all
machine
learning.
So
if.
A
B
A
A
When
really
you
look
at
how
they
were
looking
at
like
translating
images,
and
it
was
like
very
similar
to
like
you
know,
computational,
like
Neuroscience,
where
it
was
like
very
similar
and
then
like
it
kind
of
died
off
like
everyone
was
like
okay,
this
is
too
hard
and
too
expensive
and
then,
like
something
happened,
I
don't
know
at
what
point
probably
early
2000
the
group
you're
like
oh,
we
have
the
compute.
Now
we
can
probably
implement
a
lot
of
what
we've
we've
studied.
C
C
A
B
A
Like
very
similar
papers
today,
but
like
again
like
they
were
like
borderline,
you
could
barely
tell
what
images
they
were
processing
right
like
it
was
just
like
an
A
Blocky
outline.
Essentially
it's
like
okay.
We
can
kind
of
tell
the
outline
of
the
thing.
We
can't
tell
what
the
thing
is,
and
then
you
know
to
today's
time
and
seeing
a
completely
different
world.
C
We
even
tried
using
neural
networks
in
the
game.
Industry
I
was
in
the
game
industry
in
the
2000s,
and
it
didn't
really
work.
It
was
like
controlled.
It
just
ended
up
being
much
easier
to
build
a
rules-based
AI,
and
then
you
know
again
like
it
took
that
massive
massive
scaling
that
Alex
brought
to
imagenet
to
create
what
has
become
one
of
the
biggest
industries
of
the
last
10
years.
Yeah.
A
C
Yeah
I
got
a
couple
unusual,
Hobbies,
so
I'm,
a
very
Avid
pilot
I'm,
currently
flying
a
Cessna
310r
model,
the
two
other
co-founders
Evolution
IQ,
also
Pilots.
We
met
through
flying
originally
nice
I'm.
Also
a
kite
Surfer
I
had
a
very
fun
accident
in
February
I
broke
my
leg,
kite
surfing
ever
since
and
I
also
love
history.
It
gives
a
lot
of
context
and
happiness
to
understand
where
we
came
from
as
a
species
to
appreciate
what
we
have
to
say.
I
recommend
everybody
should
be
a
historian.
B
Yeah
for
me,
outside
of
work,
I
spend
a
lot
of
my
time
in
finding
new
places
to
eat
and
eating
I'm
in
New,
York,
City
and
I've
still
not
run
out
of
places
that
I
want
to
try
and
things
I
want
to
try.
But
a
lot
of
that
experiences
are
centered
around
food
and
outside
of
that
last
call
it
a
year
or
so,
a
lot
of
interest
in
crypto
and
trading.
So
nice.
A
That
would
be
interesting,
I'm
always
I'm
always
curious
like
if
it
translates
or
if
it's
worth
it.
Okay,
so
before
again
before
I,
diving
into
to
dagster
just
to
kind
of
frame
what
you
guys
do
and
how
you
kind
of
translate
it
and
needed
like
data
orchestration
like
what
does
evolution
IQ
kind
of,
do.
C
C
What
that
means
in
practice
is
we
are
an
AI
company.
We
have
a
SAS
solution
that
we
sell
to
companies
that
deal
with
bodily
injury
claims,
so
if
you're,
injured
or
disabled
there's
this
digital
trail
of
what
happened
to
you.
The
treatment
you've
gone
through
the
stage
you're
right
now
and
there's
decisions
to
be
made
very
hard
decisions
about
the
future
outcome.
What
should
happen
next?
What
treatment
should
you
be
on?
Are
you
going
to
recover?
Where
are
you
going
to
recover?
Are
you
going
to
be
able
to
work
again?
C
What
kind
of
job
can
you
do?
It's
very
complex
problem?
These
companies
are
dealing
with,
like
tens
of
thousands
of
these
claims
at
any
given
time,
they're
handled
by
a
small
group
of
people,
and
so
it
makes
total
sense
that
you
want
to
have
a
machine
that
you
compare
with
that
helps
you
understand
what
to
spend
your
time.
You
know
who,
should
you
speak
with
today?
What
should
you
you
know
what
what
are
the
next
steps
you
should
take
them
through?
C
You
know
these
disabled
injured
people,
you
know,
are
they
do
they
need,
like
additional
assistance?
Do
they
need
like
the
like
retraining
for
a
new
job,
or
perhaps
are
they
just
have
they
medically
recovered?
They
have
not
admitted
it
yet.
You
know.
Are
they
collecting,
like
an
unfair
benefit
which
is
putting
a
big
drag
on
the
cost
of
this
for
everybody
else?
B
No,
that's
actually
a
good
summary
in
a
lot
of
the
value
that
we
bring
is
in
being
able
to
help
claims
examiners
Focus
their
time
and
attention
on
claims
that
needed
the
most
because,
just
by
design
of
these
companies
and
systems-
and
there
are
not
nearly
enough
people-
then
there
are
claims
and
it's
off
by
orders
of
magnitude
we're
talking
like
25,
000,
open
claims
and
like
40
examiners,
there's
no.
They
can
look
at
everything
all
the
time
and
that's
needed
good
for
the
claimants.
B
A
Yeah
I
I
definitely
I
can
relate
to
that
because
one
of
the
first
few
companies
out
companies
I
worked
at
did
like
healthcare
kind
of
fraudulent
detection
and
that's
like
one
of
the
biggest
struggles
is
you
know
even
sometimes
when
you
you
detect
fraud,
it's
like.
Is
it
worth
adjudicating
right
because
it's
like
well,
it's
a
25
claim.
You
know
it's
like.
Maybe
maybe
it
is
fraud.
Maybe
it
is
something
but
like
it's
gonna
cost
us
a
thousand
dollars
just
to
like
you
know,
adjudicate
that
claim.
A
A
B
B
So
there
was
no
reason
to
sort
of
reinvent
the
wheel
there
and
that's
when
we
first
started
looking
at
Data,
orchestration,
Systems
I
want
to
say
well,
I
think
we
started
looking
at.
It
is
a
first
boss
pretty
early
on
Tom.
You
might
remember
like
even
in
like
late
2019
or
something,
but
the
real
pass
of
trying
to
take
it
more
earnestly
and
do
prototypes
and
trying
to
sort
of
try
to
figure
out
what
we
want
to
use
was
closer
to
late.
2020,
I
want
to
say
so.
A
B
I
think
in
the
first
few
passes,
the
four
Solutions
that
I
remember
we
looked
at
airflow
was
a
big
one.
We
looked
at
prefect
and
Daxter
and
DVC
okay,
and
when
we
airflow
I
was
also
more
familiar
with
I've
used
it.
In
my
past
life,
there's
just
a
lot
more
information
available
on
it,
but
that
also
means
that
I'm
very
familiar
with
all
the
pain
points
I
had
to
Face
by
using
airflow
that
I
would
love
to
avoid
if
I
was
doing
it.
B
The
second
time
around
the
thing
I
personally
liked
about
Daxter
was
one.
It
was
much
more
data
aware
about
what
was
orchestrating
like
it
was
built
from
for
the
particular
problem
we
were
trying
to
solve.
Airflow
was
not
like
you
can
kind
of
sort
of
use
it
for
that,
but
that's
not
what
it's
designed
for
and
that
at
some
point
makes
a
difference
in
the
tool
itself.
B
B
It's
just
that
it's
easier
to
do
it
in
the
extra
those
are
some
of
the
things
and
I
think
the,
but
also
like
not
neglecting
prefect,
is
also
very
similar
in
that
respect,
but
I
personally
found
getting
a
prototype
of
a
pipeline
up
and
running
back
then
even
much
faster
in
Daxter,
just
because
availability
of
documentation
and
the
ease
of
access
to
the
team
were
some
of
the
reasons
that
we
looked
at.
Our
talk
to
another
thing,.
C
Yeah
we
had
a
couple:
I
won't
call
them
false
starts,
but
we
invested
in
moving
from
our
custom
python
data
pipelines
to
these
apis
and
we
learned
over
and
over.
That
was
just
too
early.
We
didn't
know
enough,
and
there
was
not
enough
of
a
payback
for
us
to
make
that
change.
So
I
think
a
lot
of
companies
are
going
to
go
through
this.
C
You
know
they
start
off
as
like
sort
of
a
AI
consultancy
Shop
with
a
vision
to
become
a
platform,
and
it's
important
to
keep
that
Vision
in
sight
and
to
make
the
jump
as
soon
as
you
can
don't
wait
too
late.
You
know
don't
wait
too
long,
but
you
can't
do
too
early.
You
know,
like
you,
have
to
see
multiple
custom
data
sets
to
really
understand.
C
You
know
what
are
going
to
be
the
challenges
for
our
team
moving
forward
and
you
need
to
be
able
to
hire
kind
of
your
second
wave
of
level
Engineers,
the
first
wave
of
Engineers.
They
know
the
system
so
well.
They
don't
need
a
platform
like
they
just
know
every
line
of
code
because
they
wrote
it.
Then
you
bring
the
second
wave
and
you
realize,
like
you
know,
these
guys
are
not
moving
quite
as
fast.
They
could
use
a
better
support
structure,
and
so
you
start
to
see
I
think
from
the
from
the
people
side.
C
C
C
Sorry
I
also
point
out
that
it
was
actually.
It
was
hard
for
us
to
move
from
a
custom
python
code
base
to
what
we
have
today,
which
is
a
unified.
You
know
elt
with
the
data
warehouse.
It
eventually
took.
You
know
having
executive
sponsorship.
You
know
where
I
wrote
this
proposal
that
we
need
to
do
this
as
a
company.
You
know
we
had
to
not
just
engineering
involved,
but
you
know
we
got
everybody
involved,
so
I
had
a
one-on-one
with
all
the
leadership
in
the
company.
C
C
So
you
know
once
we
had
everybody
on
board
and
we
had
this
initiative
in
place.
Then
we
started
to
see
real
momentum
and
it
still
took
a
long
time
to
get
to
where
we
are
today,
but
it
was
a.
It
was
an
investment
that
if
we
had
not
made
it
you'd
be
a
very
different
company.
Today
I
mean
it's.
You
know
you
talk
about
what
technology
enables
and
this
investment
we
had
made
is
allowing
us
now
to
scale
to
the
next
order
of
magnitude
which
just
is
not
possible.
A
Yeah
I
think
one
interesting
thing.
You
point
out.
There
was
the
whole
like
second
wave
concept,
because
you
know,
as
someone
that
does
Consulting
I
I
often
feel
like
I
am
the
second
wave
or
fourth
wave
coming
into
someone's.
You
know
solution
that
they've
custom
built
and
that
original
engineer
leaves,
and
now
someone
needs
to
figure
it
out,
and
it's
it's
always
a
problem,
especially
if
you've
built
something
that's
very
custom.
It's
like
it's.
It's
just
hard
right
like
like
custom,
is
great
when
you've
built
it
yourself
and
it
really
sucks.
A
I
I
think
it
makes
sense
when
you're
building
like
lighter
Solutions
or
initial
solutions,
to
build
it
yourself,
but
as
you
try
to
scale
things
out,
yeah,
finding
the
right
platform,
I
I,
think
makes
sense
and
I
think
the
other
thing
I
like
that
you
brought
up
is
like
essentially
some
some
light
amount
of
like
change
management
right
like
it's
not
just
about
replacing
the
tool.
It's
about
convincing
people
that
this
is
the
right
thing
right.
C
It's
not
opposition,
it's
that
you
don't
get
any
pull.
If
you
just
propose
to
the
company
that
you're
going
to
invest
a
year
of
engineering
time
into
a
better
data
platform,
people
just
don't
they
don't
get
it
right
and
it
doesn't
work
top
down
because
people
get
busy,
they
have
other
priorities,
you
know,
and
our
company
people
have
a
lot
of
autonomy.
They
get
to
to
a
degree,
decide
what
they
want
to
work
on.
C
So
it
kind
of
takes
crafting
a
story
for
each
group
within
a
company
to
show
the
value
they
get.
It
has
to
be
real
value.
You
know
you
have
to
really
think
through
it.
You
have
to
communicate
it
well
and
once
they
can
see
it's
going
to
make
their
job
better
easier,
more
value
to
our
customers
or
shareholders.
C
Then
you
get
this
kind
of
pull
you
know,
and
then
you
get
this
like
gravitational
force
and
it
starts
to
move
forward,
but
it
takes
a
real
Champion.
Somebody
who
deeply
cares
about
this
one.
Is
it
and
it
takes
like
preparation
and,
like
you
know,
setting
up
for
battle
before
you
can
really
get
going.
C
If
you
have
an
engineer
tasked
with
this,
it's
going
to
be
very
hard
for
them
to
succeed,
because
what
they're
going
to
come
up
against
is
a
bunch
of
teams
that
have
other
priorities
and
those
priorities
will
actually
be
higher
than
the
data
platform.
If
it's
not
sold
as
the
most
important
thing
for
the
company.
So.
B
C
Know
yeah,
my
advice
is,
you
know,
find
a
champion
executive
or
not.
You
know,
have
discussions
with
every
single
person
figure
out
how
they
benefit,
and
then
you
can
get
this
thing
moving
and
then
everybody
else
just
starts
pulling
it.
You
know
towards
the
Finish.
Line
doesn't
have
to
be
the
one
person
anymore,
yeah.
A
Yeah,
no,
no
I,
think
I
think
that's
all
a
valid
and
for
anyone
out
there,
who's,
especially
technical
listening
I,
think
that's,
that's
always
important.
We
think
of,
like
the
I
think
the
project
management
component,
a
lot
like
which
is
like
how
do
I.
You
know
migrate
this
from
that
this
to
that.
But
you
know
getting
that
that
buy-in
and
that
that
change
management
is
is
also
very
crucial
because,
like
you
said,
there's
competing
priorities
and
you'll
always
get
pushed
down.
A
B
It's
not
that
I
dislike
airflow
in
and
of
itself
for
everything.
It's
just
that
for
so
okay,
so
talking
a
bit
more
concretely,
the
kind
of
pipelines
you
want
to
build
are
essentially
various
different
layers
of
air
transformation,
and
you
want
to
make
it
easy
to
both
test
and
add.
Add
to
that
air
flow
does
not
help
you
do
any
of
that.
Airflow
is
primarily
a
tool
where
you
can
easily
execute
things
at
some
Cadence
through
triggers
in
production,
which
is
very,
very
useful.
B
But
as
far
as
how
do
you
build
better
dags
or
build
better
guard
rails
around
that,
as
at
least
back
when
I
use
it,
and
when
I
tried
things
there
wasn't
anything
readily
available
or
accessible.
That
made
it
significantly
easier
for
you
to
do
that
with
airflow,
just
because
it
does
not
take
into
account
what
data
is
going
in
and
out
of
any
particular
node
on
a
dag.
B
But
if
you
are
building
a
dive
from
the
ground
up-
and
you
want
to
build
aware,
then
airflow
did
not
really
support
anything
like
that.
Like
yes,
you
could
you
could
throw
a
python
file
within
it
or
wrap
it
in
a
Docker
container
and
then
connect
all
those
things.
But
within
that
node
you
could
have
anything
and
the
other
node
was.
There
was
no
good
way
to
make
it
available
without
building
RPC
systems,
to
pass
through
a
cross
process,
information
which
was
which
is
again
doable.
It's
not
that
it's
not
doable.
A
Yeah
I
I
think,
like
the
main
issue,
that
a
lot
of
people
often
have
with
with
airflow.
Is
that,
like
I,
think
you're
insinuating
is
passing
data
right
I
remember
when
I
worked
at
Facebook
like
data
swarm,
which
is
kind
of
like
a
first
iteration
of
airflow?
That's
always
the
major
problem.
It's
like
oh
I,
want
to
make
it
like.
A
It
was
simple
things
like
I'd
be
like
I
want
to
create
a
dynamic,
dag
based
off
of
like
something
I
queried
right
and
so
I've
queried
this
data
I
have
this
data,
but
it's
it's
gone
once
the
task
is
gone
right
like
it's
like.
Okay,
now
it's
gone
so
you'd
always
have
to
like
push
it
somewhere
to
like
a
file
and
have
to
reread
it
into
a
new
task,
and-
and
that
was
always
like
just
this
extra
step
that
was
like
why'd
I-
have
to
do
this
extra
step,
rather
than
just
like
passing
it.
A
B
Yes,
exactly
because
you
can,
you
can
talk
about
how
it's
just
writing
a
file
and
passing
it.
But
then
the
moment
you
have
this
extra
concept
to
manage
when,
if,
if
you're
doing
kind
of
exploratory
work,
it
probably
matters
a
little
bit
less,
but
if
you're
doing
something
that
has
to
be
Deployable
in
production,
that
comes
with
a
significantly
larger
added
cost
of
how
to
manage
this
asset
yourself,
which,
with
extra
and
software
defined
assets
as
a
concept
indexer,
you
don't
really
have
to
build
on
your
own.
So.
A
Yeah
for
anyone
who's
wondering
Nick
Nick
is
a
co-founder
CEO
of
of
Daxter
and
then
also
one
of
the
co-founders
of
what
do
you
call
it?
Is
it
graphql
graphql?
So
you
know
he's
been
doing
data
just
more
from
my
I'd
say
a
transactional
standpoint
for
a
long
time
and
then
I
came
into
came
into
the
industries
from
a
day's
standpoint
from
Facebook
and
try
to
approach
it
from
a
different
perspective.
So,
and
he
has
a
whole
talk.
A
I
remember
seeing
him
at
the
data
Summit
doing
the
whole
talk
on
like
what
is
the
data
set,
which
is
surprising.
You
can
have
a
whole
conversation
on
that,
but
apparently
you
can,
as
you
were
kind
of
implementing
dagster.
Would
you
say
there
were
any
like
gotchas
or
tips
you'd
have
for
someone
who
maybe
was
looking
to
implement
it.
B
Yes,
several
I
think
the
the
biggest
the
biggest
thing
is
that
we
started
using
it
because
in
the
late
2020
early
2021
I
think
it
was
still
relatively
nascent
and
the
availability
of
information
on
it
outside
of
what's
been
provided
by
the
people
building.
It
was
not
that
much
but
I
just
have
a
problem
about
airflow
I
Google
it
and
it'll
be
like
50
sack
overflow
posts
about
it.
B
So
there
is
something
to
be
said
about
the
longevity
of
the
tool
that
you're
using
and
how
how
long
they've
been
around.
But
that
said
what
helped
us
a
lot
was
being
having
easily
accessible
people
in
the
people
who
are
building
Dax
or
answer
our
questions,
so
that
helped,
and
now
they
have
a
lot
more
information
available,
not
just
from
them,
but
the
people
that
are
using
it.
B
The
other
is
that,
because
airflow
has
been
around
for
a
long
time,
it's
integration
with
Cloud
deployments,
whether
it's
gcp
AWS,
whatever
your
choice
of
cloud
that
you're
using
in
in
some
way
shape
or
form,
is
already
there
so
deploying
it
from
production.
Standpoint
was
a
lot
easier,
whereas
we
had
to
do
a
lot
of
that
ourselves.
B
Those
were
the
two
main
gotchas
that,
in
the
sense
that
it's
not
like,
we
want
aware
of
those
things
before
we
went
into
it,
but
the
realization
of
how,
if
how
heavy
of
a
lift
they
would
put
on
everything
else,
was
definitely
a
realization
after
we
actually
went
into
it.
So
those
those
are
the
two
main
things
that
come
to
mind.
A
Yeah
I
I
think
it's
always
what
dagster
and
prefect
do
have
a
challenge
right
like
when
it
comes
to
like
unseeding
airflow,
it's
like
when
it
comes
down
to
airflow.
It's
just
been
around
longer.
If
you
have
a
problem,
you
Google
it
it's
likely
there
somewhere,
like
you
said,
like
you,
throw
up
mwaa,
Cloud,
composer
astronomer.
There
are.
There
are
versions
that
are
already
managed
that
exist
in
that
you
know,
simplify
things
even
further,
like
you
said,
I
think
Daxter
now
has
a
a
cloud
version.
That's
out
of
beta
I,
think.
A
A
B
B
But
the
existing
set
of
tooling
that's
available
for
machine
learning,
pipeline
creation
outside
of
this
concept,
even
cycle
and
pipelines
or
tfx,
or
something
like
that
is
so
much
more
mature
and
rich
in
in
features
that
there
is
like
a
gap
to
be
filled
in
using
daxer
and
the
at
least.
As
far
as
I'm
aware,
the
connectivity
between
Daxter
and
those
ml
toolings
is
not
as
mature
or
does
not
exist.
B
A
No
I
I'd
love
to
go
into
it,
just
because
I
one
I'm
sure
you're,
probably
already
telling
Nick
and
his
team
about
it.
But
you
know
just
because
I
I
think
like
it
does
bring
that
whole
thing.
Full
circle
right
like
it's,
not
just
the
data
engineering
aspect
of
the
pipeline
like
at
Facebook.
There
are
a
lot
of
you
know:
data
scientists
and
NM
and
ml
Engineers,
who
are
using
data
swarm
to
do
their
models
right
so
and
there's
other
companies
that
are
using
airflow
to
like
manage
their
models.
B
Yeah
absolutely
so
kind
of
here
is
here
are
some
of
the
things
that,
over
the
last
10
years
of
doing
machine
learning
that
I
learned
the
hard
way,
and
one
of
one
of
the
biggest
challenges
is
like
thinking
about
a
model
as
just
the
The
Learning
System.
That
is
an
architecture,
be
it
like
simple
stuff,
like
regression
and
svms,
to
like
more
complicated
neural
network
calculations.
That's
not
it!
That's
not
just
the
model,
there's
a
whole
host
of
feature
engineering
and
cleaning
data
cleaning
that
goes
into
it.
B
There
are
series
of
Transformations.
Some
are
some
are
all
transformations.
Some
are
learned,
Transformations
even
and
the
biggest
mistakes
that
get
made
in
building
machine
learning.
Pipelines
like
this
is
forgetting
to
correctly
apply
the
same
Transformations
that
you
did
while
training
your
model
and
testing
your
model
in
the
production
system.
That
you're
doing
it's
very,
very
easy
to
make
that
mistake,
and-
and
it
should,
whenever
possible,
you
should
not
rely
on
processes
for
doing
that
or
having
the
engineers
remember
to
do
that.
B
If
it
can
be
automated
away
or
just
that's
something
you
don't
even
have
to
think
about,
then
that's
that's.
Even
better,
so
what's
something
like
scikit-learn
pipeline
might
allow
you
to
do.
Is
you
have
like
four
or
five
feature
Transformers
you
have
a
model
at
the
end
of
it
that
entire
pipeline
can
can
be
trained
and
saved
when
I
say
saved,
I
mean
serialized,
with
both
both
the
code
and
the
Learned
parameters
to
disk.
B
So
when
you
load
it
and
call
model.predict
on
whatever
data
you're
passing
to
it,
you
don't
have
to
worry
about
making
sure
that
all
the
Transformations
that
I
did
during
training
my
set
are
also
being
applied
to
the
the
system
that
I'm
using
for
production,
because
that
will
be
done
by
the
pipeline,
because
all
those
all
that
information
is
already
saved.
You
can't
make
a
mistake
there
because
you're
not
trying
to
remember
to
make
sure
all
the
modules
are
properly
imported
or
the
functions
are
called,
or
things
like
that,
like
that.
A
B
Hugely
important
piece
that
that
is
a
piece
of
functionality
that
you
can't
forego,
because
you're
just
going
to
introduce
a
lot
more
complexity
in
making
those
mistakes
and
solving
for
them.
So
that's
kind
of
one
very
big
example
of
what
some
things
psychic
learn
allows
you
to
do
or
even
and
having
to
use
DFX
in
a
while
in
a
couple
years
ago,
but
I'm
pretty
sure
it
allows
you
to
do
that
as
well.
B
And
that's
that's
an
understated
feature
that
is
not
often
talked
about
and
as
far
as
I
know,
not
a
not
a
clear
way
to
do
that.
In
Daxter,
though,
there
are
the
Core
Concepts
and
fundamentals
that
are
needed
to
fully
support
that
exist
right.
Essentially,
what
you
need
is
being
able
to
run
same
nodes
of
a
dag
with
same
configurations
in
different
modes
such
that
the
node
behaves
slightly
differently
in
those
two
modes
and
those
need
to
be
user
definable
modes.
So
you
can
have
a
train
motor
test
mode,
an
inference
mode.
B
Whatever
you
want,
and
then
you
want
to
be
able
to
serialize
both
the
how
the
diet
connects
to
each
other.
But
the
configurations,
as
well
as
any
and
all
learned
parameters
which
can
be
software
defined
assets
for
that
particular
node
as
a
part
of
it.
So
it's
doable
for
sure,
but
it
will
need
some
more
custom
work
on
top
from
us.
B
If
we
sort
of
start
doing
that-
and
that's
I
think
that's
the
biggest
feature,
that's
missing
for
for
me
and
without
that
I
like
I'm,
not
willing
to
give
up
that
feature
to
get
it.
But
the
thing
that
I
want
from
Baxter
that
scikit-learn
doesn't
have
is
because
you
have
the
idea
of
software-defined
assets.
B
You
can
relatively
easily
set
up
a
pipeline
to
be
run
from
the
middle
without
building
custom
versioning
for
it,
because
if
you
have
every
node
or
every
node
that
you
care
about
with
an
asset
that
it
throws
out,
you
can
easily
have
it
be
dependent
on
both
the
code.
That
was
run
as
well
as
the
input
that
it
got
and
if
you
have
that
you
run
a
pipeline
for
whatever
reason
fails
them.
B
It
doesn't
matter
why,
when
you
click
run
the
next
time,
if
everything,
if
nothing
in
the
code
or
the
input,
is
changing,
it
won't
recompute,
you
don't,
you
can
set
it
up
to
not
really
compute
it
and
only
run
after
after
that
fact,
that's
hugely
beneficial,
especially
when
you
are
kind
of
in
the
mode
of
building
a
new
pipeline
testing.
Some
things
things
are
breaking
and
you
just
want
it.
A
C
The
things
there's
one
more
interesting
topic
I
want
to
ask
you
guys
about
it's
the
adoption
curve
so
orchestration
framework.
When
you
have
a
25
engineering
team,
they
don't
actually
need
to
use
it
right.
They
can
just
keep
hacking
around
it
and
just
like
you
have
in
the
consumer
Market
you
have
these
personalities
of
like
early
adopters,
later
adopters,
laggards,
I
think
the
same
exists
in
engineering
and
most
Engineers
are
not
early
adopters.
They
don't
want
to
jump
into
the
new
API
and
start
using
that.
If
there's
something
that
works
right,
they
have
their
like.
C
You
know
they
have
their
tasks
better,
they
have
to
fix
bugs
and
they
have
to
like
run
features
and
launch
new
clients.
So
how
do
you
get
an
entire
team
that
is
already
like
overloaded
with
work
start
adopting
and
migrating
to
this
thing
that
they
can't
quite
see
the
benefit
of
yet
we
so
we
solved
this,
but
it
wasn't
easy
I'll
be
curious
to
hear
for
both
of
you
guys.
You
know
how
do
you
see
Quran
looking
back
at?
C
You
know
how
we
did
it
and
then
you
know
how
you've
seen
this
done
to
get
through
this
challenge
like
how
can
other
companies
plan
for
getting
this
into
the
state?
They
have
this
orchestration
Pipeline
and
it's
great,
but
nobody
really
wants
to
use
it,
and
you
can't
force
people
to
use
it.
You
have
to
find
some
other
method
to
gain
adoption.
B
A
Is
always
again,
this
is
like
the
hard
part
right.
Technology
is
easy.
People
are
hard
and
like
what
works,
it
doesn't
even
always
work
the
same
way
in
every
company,
I
think
to
some
degree,
what's
like,
let's
say
like
what's
worked
with
like
at
Facebook,
Facebook
essentially
allows
you
to
use
any
any
whatever
you
want
to
use
to
work
with.
Do
it
right
like
if
you
want
to
use
vs
code
USBS
code,
if
you
want
to
use
something
else,
use
that
and
what
they've
done?
A
A
You
want
to
use
something
else
great,
but
also
all
that
other
stuff's
not
really
fully
supported
so
I
think
that
that's
one
interesting
way
it's
like
just
to
kind
of
like
force
people
through
through
no
other
choice,
I
I,
think
also
it's
not
to
be
like
a
dictator,
but
I
think
there
is
something
to
be
said
about,
like
you
have
to
pick
a
standard
and
letting
everyone
know
like
okay.
This
is
the
standard.
This
is
what
we're
gonna
do.
This
is
how
we're
gonna
do
it
moving
forward.
A
I
think
someone
generally
has
to
make
that
call
regardless
there's
a
there's,
a
chapter
in
the
mythical
man
month,
where
they
talk
about
like
building
Cathedrals
and
how
like,
in
the
past,
like
they
were
like
some
cathedrals
were
built
over.
You
know
centuries,
but
kept
like
the
same
basis
over
those
centuries,
despite
like
different
Architects
and
people
building
over
time,
where
others
didn't
and
you
could
tell
and
so
I
think
there
is
something
too
like
setting
standards
and
saying
like
hey.
This
is
just
what
our
team's
gonna
do.
A
These
are
the
reasons
why
you
know
you're
always
gonna
have
challenges.
I
always
come
up
with
people
who,
for
example,
have
the
archetype
of
like
I'm,
an
open
source
person
right
like
they're,
always
going
to
want
to
use
open
source
technology.
You
know
regardless
and
I.
Think
sometimes
you
just
like
I,
don't
know
someone
said
to
put
your
foot
down
and
be
like
I
get
it.
This
is
what
our
team's
gonna
do.
This
is
what's
going
this
part
of
our
team.
A
B
Yeah
I
think
I
think
the
thing
what
you
said,
sort
of
reminded
me
of
the
balance
between
autonomy
and
and
Direction
is
hard
to
achieve
and
into
polaroxically
also
depends
on
your
business
setup
and
how
you
compute
Roi,
based
on
what
an
individual
engineer
produces
from
that
standpoint.
So
and
I
think
a
lot.
B
If
you're
operating
at
Facebook
scale
yeah,
you
can
afford
to
say
a
lot
more
of
like
hey
use,
whatever
you
like,
as
long
as
you're
producing
value
and
having
impact
so
anyway,
but
to
go
back
to
what
sort
of
form
was
asking
I
think
the
way
we
did.
It
was
when
om
sort
of
started
talking
about
their
Sac
2.0
at
our
company
and
evangelizing
it
and
starting
to
talk
about
this
idea.
B
B
B
B
really
helped
in
getting
over
that
hump
of
doing
that,
so
you
have
to
in
a
small
in
a
small
company
like
ours
at
least
you
have
to
treat
everybody
as
a
owner
because
they
are,
and
that
means
that
you
provide
them
with
the
right
information,
how
you're
thinking
about
it
and
if
they
actually
disagree
with
you,
that's
good,
because
you
probably
learned
something
and
overlooked
something,
and
you
incorporate
that
into
the
path.
It
might
be
a
slightly
longer
path,
but
you'll
be
on
a
better
path.
There.
B
C
Yeah
you
know
I,
remember
a
moment,
though
this
was
well
into
Data
stack
too,
and
Daxter
was
being
rolled
out
and
I
checked
with
a
couple
engineers
and
they're
just
kind
of
complaining
about
it.
You
know
they're,
like
they,
don't
really
see
the
benefit.
You
know
we
just
think
we
have
problems
with
it.
It's
not
being
used
correctly
and
I.
Just
couldn't
figure
out
what
to
do.
You
know
like
we
have
gone
so
far,
and
yet
we
just
could
not
get
adoption
from
our
engineering
team.
C
I
talked
to
him
about
this,
and
I
told
him
that
this
is
happening
like
there's,
not
an
option,
and
he
took
it
on
himself
to
spend
time
with
every
engineer
personally
and
like
sit
with
them
and
talk
to
them
and
like
look
at
their
code
and
show
them
how
much
better
life
will
be
with
Daxter,
so
I
think
for
us.
It
took
having
that
inside
engineer,
sponsor
who
wasn't
a
lead,
they're,
not
a
manager.
You
know
they
were
just
an
expert
and
they're
passionate
about
this,
and
they
understood
it
really.
C
Well
so
I
think
that
intersection
of
like
passion
and
deep
technical
expertise
and
respect
on
the
team
is
what
it
took
eventually
to
get
people
to
really
like
realize.
Like
you
know,
I
can
do
this.
It
is
going
to
provide
Great
Value
to
the
company.
It's
going
to
make
my
life
better
and
after
that
I
think
things
were
like
going
really
well
I.
Don't
think
that
I
had
any
more
concerns
about
whether
or
not
we're
gonna
get
to
the
end
of
this
road
yeah.
A
Yeah
I
think
that
sounds
like
and
I.
Think
I'm
gonna
cover
your
team,
your
face,
oh!
No,
it
kind
of
moves
you
up,
Dan
kind
of
references.
This
right,
like
you,
gotta,
find
some
Champions
at
the
ground
level
to
kind
of
trying
to
kind
of
push
it,
and
there
were
definitely
cases
even
at
Facebook
like
where
someone
would
be
like
it
felt
like
every
quarter.
There
was
someone
pushing
some
new
thing
like
like.
A
Oh
now,
we've
got
column
level
security,
it's
like
oh
gosh,
now,
I
gotta
go
back
through
all
my
pipelines
and
add
and
call
muscle
security,
and
then
there
would
always
be
someone
who's
like
Hey.
We're
gonna.
Do
this
we're
gonna
make
it
happen,
so
so
I
I
guess
that
that
does
kind
of
kind
of
ring.
True
it's
like
yeah.
You
usually
have
to
find
someone.
That's
gonna
like
own.
A
It
and
kind
of
also
be
there
and
be
willing
to
like
help
those
last
few,
because
there's
always
gonna,
be
that,
like
last
20
of
people
for
various
reasons,
right
like
like
you
said,
they're
busy,
they
don't
feel
like
they
see
the
value
adopt
and
you're
gonna
have
to
win
those
over
the
other.
People
will
be
easy
right
like
it's
like.
Okay,
you
get
the
early
adopters
they'll
be
fine.
C
Think
the
the
other
issue
is
that
Engineers
have
a
tendency
to
build
processes
which
is
great
right,
documentation
create
examples.
Sometimes
you
just
need
to
throw
it
all
away
and
just
like
sit
with
the
other
person
and
talk
to
them
and
like
connect
to
the
human
Love.
Sometimes
that's
the
most
efficient
way
to
solve
impossible
problems
right
like
in
the
end,
we're
also
people.
A
C
So
I
have
a
specific
view
on
this.
You
know
we
create
and
store
so
much
data
every
day.
Now
the
question
is
how
much
of
that
data
is
being
used
today
for
advanced
machine
learning,
it's
hard
to
measure
it,
but
I
think
it's
safe
to
say
it's
like
less
than
a
percent,
maybe
less
than
a
tenth
of
a
percent,
certainly
like
in
our
industry.
C
It's
very,
very
low,
I
think
over
the
next
five
to
ten
years,
we're
going
to
reach,
like
you
know,
double
digits
like
we'll,
actually
be
utilizing
more
than
10
of
the
data
in
the
world
for
predictions
and
the
impacts
of
that
are
impossible
to
predict.
But
you
know
it's
going
to
be
it's
going
to
change.
You
know
everything,
so
you
know
I
think
people
should
think
about
this.
Where
are
we
at
today
in
terms
of
their
usage
for
Predictive
Analytics?
C
What's
the
opportunity
out
there,
you
know
how
big
is
the
world
of
data
and
how
fast
we're
moving
and
upgrading
to
it.
So
I
think
we're
just
at
the
very
very
onset
of
the
era
of
AI
the
era
of
light
guiding
humans
with
AI
and
we're
going
to
see
some
huge
huge
changes
still
coming.
You
know,
I,
don't
think
that
we
are
anywhere
near
T
capacity,
yet.
B
These
things
have
been
having
more
front
and
center
for
them
and
over
a
period
of
time
they
will
become
sort
of
a
common
place
in
the
background
of
the
decision
making,
because
they
will
start
to
see
the
benefit
of
leveraging
it
from
the
get-go.
So
it
won't
be
a
conversation
about
data-driven
decision
making
versus
non-dairy
driven
decision
again.
Really
it's
just
decision
making
and
yeah
how
you
make
it
is
using
using
data
I,
don't
think,
at
least
in
my
observations
and
my
experiences
in
from
large
and
small
companies.
B
A
A
C
I
was
going
to
add
on
to
what
cron
said
the
other
emerging
change
that
we
see
happening,
and
we
see
this
happening
in
our
company.
We
think
it's
going
to
happen
throughout
the
rest
of
the
you
know,
kind
of
data
world
is
we
started
off
with
AI
being
used
for
a
lot
of
automation,
tasks.
You
know
like
label
this
image.
C
You
know,
recommend
this
song
like
they're,
basically
replacing
what
a
human
would
do
in
the
past.
What
we're
doing
is
a
little
bit
different
right,
we're
saying,
like
you
know,
human
has
a
job
to
do
they're
doing
something
complex
and
they
benefit
from
having
an
assistant
having
like
this
expert,
helping
them
with
part
of
their
tasks,
I
think
in
the
future,
every
almost
every
job
in
the
world
is
going
to
have
some
kind
of
an
AI
assistant
working
with
the
human
right.
C
Just
it
just
makes
sense
that
you
don't
want
to
be
working
alone,
but
you
don't
want
to
have
you
know
computers
completely,
replacing
what
you
do,
they're
not
going
to
be
as
good
at
some
of
the
high
level
decision
making
and
task
setting.
So
you'll
start
to
see
a
lot
of
investment
in
enabling
people
to
do
their
jobs
better.
A
C
If
I
experiment
to
like
think
of
the
world
as
a
giant,
hard
drive
and
a
giant,
you
know
data
pipe,
and
you
can
probably
estimate
measure
how
much
total
data
exists
in
the
world
today
and
how
much
he's
kind
of
created
or
transferred
on
a
daily
basis.
Then
look
at
all
of
the
people,
all
these
Engineers
that
have
machine
learning.
C
C
How
many
people
today
and
how
many
companies
today
are
the
importance
of
analytics
I
think
they
give
you
a
rough
sense
of
just
how
far
we
still
have
to
go
and
I
think
the
number
is
very
large,
like
I
said,
I
think
you'll
come
up
with
like
a
fraction
of
a
percent
of
both
the
data
side
and
the
people
and
the
company
side
that
is
in
the
technology
today
in
the
future.
I
think
we're
going
to
get
close
to
100
just
like
today.
How
many
companies
are
not
using
some
kind
of
a
digital?
C
You
know
storage
like
even
if
you're
a
your
contract
for
doing
like
local
plumbing
right.
You
have
some
kind
of
a
customer
database
and
you
have
like
you
would
communicate
with
your
clients
and
you're
using
you
know,
accounting
software
like
it's.
It
would
be
crazy
today
to
do
everything
on
paper.
I
think.
In
the
same
way,
it's
going
to
feel
crazy
in
I.
C
Don't
know
it's
going
to
be
5,
10
or
15
years
to
you
know,
be
doing
some
kind
of
a
job
of
and
not
be
benefiting
from
a
predictive
animal
system
that
is
consuming
Mass
amounts
of
data
from
the
industry
in
the
world
in
real
time
and
helping
you
like,
basically,
you
know,
behave
like
a
superhuman
organism.
Yeah.
B
Not
so
much
thrown
around,
but
maybe
more
consolidation
around
things
that
are
not
don't
need
to
be
called
out
of
separate
terms,
but
they
just
add
more
I.
Think
if
you
the
more
the
more
new
terms
you
keep
adding
in
a
domain,
it
almost
acts
as
a
barrier
for
new
people
to
enter
it
easily,
so
I
I
wish.
We
would
stick
to
simple
Concepts
that
Define
things
and
I
don't
think.
B
A
Okay,
now
we
need
something
else
right
like
now
that
the
new
Fascination
is
right.
Like
data
contracts
like
we
need
data
contracts
now,
which
aren't
necessarily
new
but
like
the
the
like
need
of
them,
is
like
being
established
and
stuff
to
say
that
we
don't
need
them
or
we
do
need
them.
It's
just
like
another
thing.
That's
now
like
okay,.
C
Service
to
humanity
that
that
science
has
decided
to
use
any
completely
new
language.
All
these
terms
that
don't
exist
for
engineers
right
I,
think
it's
made
data
science
and
machine
learning,
much
less
accessible
to
programmers.
It
feels
to
them
like
it's.
This
opaque
archaic
like
new
super
science,
and
it's
not
I
mean
it's
math
and
it's
data.
You
know
I
I,
wish
that
this
didn't
happen.
I
wish
that
we
didn't
come
up
with
a
bunch
of
new
terms
to
describe
things
that
are
pretty
logical.
C
A
B
A
Yeah
I
I
think
I
agree.
It's
like
at
the
end
of
the
day,
like
that's
I,
think
that's
why
a
lot
of
people
do
adopt
airflow
nowadays
and
when
I
did
do
kind
of
like
a
deep
dive
into
airflow
and
I
asked
people
like
where
they
started.
A
lot
of
people
said
airflow,
which
I
was
honestly
expecting
more
people
to
say,
like
KRON
and
like
batch
Scripts,
because
I'm
like
yeah
Bash.
That's
where
like
that's
where
everyone
starts
like
some
Python
scripts
on
KRON
or
some.
A
C
B
B
A
Know
it's
like
because
or
or
use
like
land
of
functions
or
or
Cloud
functions
with
like
I,
don't
know
what
the
gcp
version
of
like
basically
schedule!
Cloud
scheduler
is
I.
Think
it's
just
Cloud
scheduler
like
that's
all.
You
need
the
problem
with
like
mwa
and
Cloud
composer.
Is
they
actually
have
pretty
expensive,
pretty
pretty
quickly
because
they're
always
running
so,
if
you're,
really,
if
you're
just
doing
a
simple
daily
job,
it's
like
just
just
spin
up
something
daily,
that's
serverless
and
pay
for
that.
B
Yeah
I
think
for
for
two
people
company,
that's
just
like.
Let's
make
it
more
concrete
like
if
I
was
starting,
my
own
company
today,
Tom
you
and
I
were
starting.
Our
own
company,
like
I,
would
pick
the
tool
that
I'm
most
familiar
with
like
because
I
don't
want
to
spend
my
time
on
tooling.
I
want
to
spend
my
time
on
building
the
thing
that
would
make
this
into
a
company
and
build
a
product
right.
B
So
if
I
happen
to
be
more
familiar
with
cron
and
Python,
and
spinning
up
gcp
VMS,
but
not
with
Lambda
functions,
I'm
not
going
to
do
that,
and
this
is
a
real
cost
benefit
for
it.
If
I
already
have
used
airflow
for
the
last
five
years
and
I'm
familiar
with
gcp
and
composer,
that's
what
I'm
going
to
spin
up
so
pick
the
thing
that
you're
most
familiar
with,
if
you're
familiar
with
something
and
most
people
are
familiar
with
something,
don't
think
of
it
as
a
data
orchestration
tool.
B
A
B
A
And
a
pivot
table
right
like
it's
like
look
you're
just
needing
to
answer
how
many
customers
you
have
you
don't
even
need
a
data
warehouse.
You
have
two
people,
you
know
it
might
be
Overkill
to
like
code,
something
again
it
depends
if
you're
running
this
report
for
a
weekly
basis
or
you
know
daily
basis,
but
if
you're
doing
it
once
a
month,
it's
like
yeah,
it
might
make
sense
to
just
pull
up.
A
Let
me
think
there's
a
lot
of
comments
so
like,
for
example,
someone
commented
about
airflow
airflow
is
like
Jenkins
Jenkins
the
ubiquitous,
but
there
are
shortcomings
that
other
newcomers
solve
for
I
think
it's
never
resolved
it.
They
truly
tried
with
blue
ocean,
but
it's
still
not
widely
used.
I
I've
actually
seen
a
few
people
use
Jenkins,
similar
to
like
airflow
like
they'll,
like
have
they'll
code
python
tasks
on
it
and
then,
like
you,
know,
push
it
through
Jenkins
and
then
run
it
through
Jenkins
and
I've.
B
I've
seen
similar
things
happen
as
well
like
where
you
you
might
have
written
software
in
a
lot
of
cicd
pipelines
and
you're
thinking.
Oh
I
just
need
to
run
a
couple
of
things
one
after
the
other
or
in
some
kind
of
for
that
fashion.
I
know
Jenkins,
which
is
not
necessarily
the
worst
idea
to
begin
with.
The
thing
that
I
struggle
with
is
the
same
ideas
have
different
sort
of
names
and
Concepts
that
don't
really
resonate
well
and
some
Concepts
overlap.
B
C
C
B
Quite
quite
a
lot
right
because
the
the
stronger,
the
community,
the
more
information
there
is
out
there,
the
more
quickly
things
get
fixed,
the
more
quickly
you
new,
Concepts
and
ideas
get
added
to
it.
The
faster
it
grows,
so
yeah
I
mean
if
you,
if
you
want
to
use
something
for
not
just
the
next
couple
months
and
more,
have
some
kind
of
longevity
in
mind.
B
It
does
matter
a
lot.
So
even
when
you're
pick
like
or
I
do
this
when
I'm
taking
like
a
small
tool
or
a
python
library,
that
does
something
the
first
thing.
I
check
is,
if
it's
being
actively
maintained
or
not
like,
if
it
has,
there's
no
commit
for
a
year
and
a
half,
probably
gonna,
go
for
something
else.
C
Yeah
yeah
and
if
the
company
behind
the
commercial
company
doesn't
succeed,
which
can
happen
for
various
reasons
right
if
it's
open
source,
you
know
that
it's
going
to
be
out
there
for
you,
and
at
least
for
the
next
couple
of
years
you
know
you're
not
going
to
have
to
jump
and
migrate
to
something
new.
So
it
is
an
important
decision,
especially
for
new
technologies
that
have
not
don't
have
like
a
billion
dollar
market
cap
behind
them.
Yet.
A
Yeah
yeah
I
I
think
like
it
goes
with
whether
it's
open
source
or
non-open
Source,
obviously
I
think
with
open
source,
even
more
so
but
like
even
with
not
open
source.
The
larger
the
community,
like
the
more
the
more
reason
to
use
it
right,
because
it's
just
like
when
you
do
run
into
those
problems,
because
they're
they're
always
problems
right
like
no
matter
the
solution,
that's
being
built
out
there.
A
There
are
shortcomings
and
you
need
workarounds
and
it's
nice
to
have
those
workarounds
documented
somewhere,
like
the
tablet,
will
work
around
for
making
a
donut
chart.
I
don't
know
if
anyone
knows
that,
but
it's
it's
there.
It's
very
funny
to
me
and
it's
like
yeah
it's
there,
so
you
can
Google
it
and
find
it
and
find
a
million
different
people
writing
an
article
or
making
a
video
on
it.
A
So
personally,
there's
I
adopt
things
if
they're
being
used,
because
otherwise,
like
the
amount
of
documentation
or
like
troubleshooting
you're
gonna
have
to
do
is
just
like
exponentially
more
because
you're
just
gonna
be
spending
time
like
testing
out
different
things
like
is
it
dot
shift
left?
Is
it
dot
substring?
Is
it
dot
left?
Is
it?
What
is
it?
What
is
the
the
thing?
So
just
because
this
is
here,
does
anyone
have
any
thoughts
on
data
mesh,
since
it's
puzzling,
some
people.
A
A
Yeah
I
mean
like
I'm,
not
gonna,
like
sit
here
and
I
I,
just
end
up
finding
a
doc
like
a
dictionary
definition
but
I.
The
interesting
thing
to
me
about
data
match
is
that
like
in
some
ways,
Facebook
like
from
what
I've
heard
about
like
datamesh
has
implemented
one
essentially
right
like
you
do
it
without
thinking
about
it.
A
What
is
this?
Oh,
exactly
the
point
so
many
terms
up:
yeah
yeah,
it's
just
I-
was
kind
of
an
interesting
like
challenge
recently
because,
like
people
realize
that,
if
you,
if
you
are
to
coin
a
new
term,
there's
a
lot
of
value
there
right
if
you're,
if
you're
the
coiner,
if
your
product
is
connected
to
it,
either
you
get
Consulting
because
of
it
or
you
get
more
clients
or
more
customers
because
of
it.
A
A
C
Yeah
I'll,
first
of
all,
I
want
to
put
in
a
plug
for
the
Dexter
team,
phenomenal
team.
You
should
look
to
see
who
works
at
the
company
whose
API
gonna
use
so
I,
always
look
not
just
at
the
GitHub
usage.
I
looked
at
the
LinkedIn
profiles
of
the
engineers
at
the
company
and
I
know
Nick
personally,
but
I
can
you
know.
Do
this
exercise?
Look
for
yourself
you're
going
to
see
they
have
like
some
of
the
messages
in
the
world.
C
I
do
this
for
any
company
you
work
with
I
will
also
put
in
a
plug
for
us.
We
have
one
of
the
best
ax
news
in
the
world,
one
of
the
best
data
teams,
one
of
the
best
application
teams.
We
are
hiring.
So
if
there's
anybody
out
there
looking
to
join
a
hot
AI
startup,
please
retract
me
reach
out
to
Quran
connect
with
us
on
LinkedIn,
even
it's
just
going
to
network
with
us.
If
you
want
to
get
a
conversation
started,
you
know
throw
ideas
back
and
forth.
B
Yeah,
so
regarding
Baxter,
the
thing
that
sold
me
also,
one
of
on
using
Daxter,
was
some
of
the
talks
that
Nick
gave
and
just
the
clarity
of
thought
process
around.
How
he's
thinking
about
the
problem
space
gives
me
the
confidence
of
that
the
the
longevity
of
the
store
is
going
to
be
great
and
investing
in
it
early
means
that
we
can
help
shape
and
grow
it
in
the
direction
that
our
needs
are,
and
we
definitely
get
that
access
early
on.
B
So
that's
definitely
there
and
definitely
look
at
the
theme
building
it,
whether
it's
an
open
source
Community
or
it's
a
company,
it's
very,
very
useful
to
know
who
who
the
people
are
behind
it
and
having
some
confidence
in
their
ability.
Plugging
for
ourselves
also
we're
hiring
a
lot
ml.
Engineers
data
Engineers,
analytics
Engineers
data
analysts,
devops
Engineers,
full
stack
engineers,
back-end
engineers.
B
A
Of
that
was
a
lot
of
open
roles.
Maybe
maybe
I
missed
that?
No
that's
exciting.
So,
if
anyone's
complaining
about
not
finding
work,
they're
they're,
there's
work,
yeah
well,
I
appreciate
your
guys's
time
and
I'm
gonna
hit
and
live
here
in
a
second.
But
thanks.
Everyone
watching
and
I
will
see
you
guys
next
week,
so
I'm
gonna
hit
n
live.