►
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
So
the
inaugural
meeting
of
the
applied
machine
learning
weekly
group,
which
is
which
is
great
so
good
stuff
and
so
welcome
again
out
looking
through
the
agenda,
welcome
again
alexander
to
gitlab
and
glad
everybody
else
will
be
helping,
including
mihao
and
taylor,
who
is
not
on
today
and
others.
So
we
have
a
team
page,
it's
linked
in
the
agenda,
which
is
good
stuff.
A
The
first
epic,
which
we
started
talking
a
little
bit
yesterday
with
the
with
the
infrastructure
team,
which
was
a
great
discussion
with
them,
we'll
really
get
started
on
it.
After
alexandria
completes
you
know,
onboarding.
We
want
to
make
sure
we
give
you
the
time
alexander
to
do
it
and
you've
only
been
doing
that
for
a
week
and
a
half,
and
it
takes
at
least
two
weeks.
You
don't
definitely
don't
feel
the
pressure
to
start
working
on
it
in
earnest
until
after
you're,
primarily
done
with
onboarding.
A
The
on
that
you
had
some
technical
questions
that
we
need
to
answer
that
you
started
discussing
yesterday.
Alexander
and
you're
gonna
create
an
issue
for
those,
so
we
can
maybe
track
it
transparently
in
an
issue.
Some
of
those
things.
A
A
Do
you
want
alexander,
do
you
want
to
discuss
any
of
those
today?
Do
you
zoom
or
do
you
want
to
wait
and
we
do
it
async
later
either
way?
It's
fine.
B
Doesn't
matter
there
are
a
lot
of
questions
to
discuss
so,
for
instance,
the
main
question
for
me
is
how
to
replace
some
of
the
components-
and
you
know
I
think
mon
she
saw
she
watched
our
video
yesterday,
so
she
proposed
some
things.
B
A
A
Dataflow,
okay,
I'll
put
in
I'll
put
in
an
item
number
item
number
two,
okay,
so
so
one
thing
as
we
think
about
our
proof
of
concept:
the
is
to
implement
it
on
review.
Basically,
as
is
as
much
as
we
can,
which
we
know,
we
know
we
won't
be
able
to
do
that.
But
you
know
because
we're
moving
it
to
gcp
from
azure
milestone.
2
is
about
integrating
with
the
product
for
com
customers,
not
for
self-hosted
customers.
A
When
we
talk
about
self-hosted
customers,
running
machine
ml,
workflows
is
very
da,
is
very
intensive
that
so
we
may
make
decisions
that,
but
if
we
wanted
to
be
able
to
run
on
cell
phone
customers,
we
want
to
keep
that
in
mind
when
we
make
decisions
and
there's
this
whole
discussion
of
other
ways
of
doing
things.
So
I
don't
know
taylor
if
you
have
any
thoughts
on.
A
Is
it
okay
for
us
to
make
assumptions
that
you
know?
We
won't
be
installing
all
these
components
for
self-hosted
customers
and
they
have
an
alternate
solution,
or
should
we
be
keeping
in
mind
the
self-hosted
customers
as
we
come
up
with
their.com
solution?
You
know
there's
pros
and
cons
of
both
approaches
but
curious
about
your
thoughts
at
this
point.
C
Yes,
I
mean
we're
in
pretty
uncharted
territory
here.
I
think,
keep
in
mind
that
when
we
talk
about
self-hosted
customers,
they're
running
self-hosted
or
they're
running
their
own
runners,
and
so
one
thought
is
maybe
this
is
just
a
special
runner
type
that
they
need
to
be
able
to
configure
to
and
make
available
to
power.
This.
A
Great
so,
and
the
idea
there
in
that
kind
of
self-hosting
would
be.
We
extract
the
data
on
the
self-hosted
instance
and
then,
if
customers
want
us
to
do
this,
then
we
send
it
up
to
our
cloud
with
the
customer's
configuration
and
permission
run
the
ml
there
and
then
take
the
results
and
send
it
back
to
the
self-hosted.
A
So
I
guess
any
components
we'd
need
to
do.
The
extraction
would
need
to
run
on
self-hosted
and
be
appropriate
for
that,
but
the
all
everything
else
would
be
it'd
be
okay
to
be
in
the
cloud.
Perhaps
so
it's
a
good
way
to
think
about
it
and
it's
a
much
smaller
scope
challenge
for
us
to
support
self-hosted
attention.
It's
kind
of
cool
so
also
welcome
me
how
to
the
team
you
know
temporarily
transitioning
in
mid-june,
you
know
maybe
permanently
you're
gonna
make
that
decision.
We've
got
a
transition
issue
later.
A
You
know
you,
and
I
both
you
know
this
is
something
you
like
more
than
you
than
what
you
were
doing
previously
or
less,
and
if
so,
if
you
like
it
less,
you
know,
go
back
to
the
other
team
and,
if
not
stay
on
this
team
permanently
so
again
welcome
to
the
team,
I'm
looking
forward
to
we're
looking
forward
to
leveraging
your
background
in
ml
as
well.
D
All
right
thanks,
I
did
write
some
questions
in
the
agenda,
so
I'd
like
to
like
vocalize
them,
but.
C
D
Jay
went
out
of
the
window
alexander
reminded
me
that
no
longer
we
no
longer
depend
on
the
neo4j.
So
that's
not
a
problem.
I've
been
wondering,
since
you
explicitly
mentioned
that
we
use
mongodb
in
on
review.
Did
you
consider
using
a
json
bcom
in
postgres?
This
would
be.
This
would
mean
essentially
less
work
for
our
infrastructure
department,
especially
when
it
comes
to
omnibus
installations.
So
we
don't.
D
We
wouldn't
have
to
ship
mongodb
to
our
customers
if
we
decide
to
integrate
it
in
omnibus
installation
and
that
should
solve
our
the
problem
of
like
maybe
not
a
problem.
This
is
not.
This
is
not
a
problem,
this
additional
moving
component.
So
this
is
something
we
have
to
consider
carefully.
D
I
am
not
sure
what
would
be
the
expected
load
of
the
database
when
it
comes
to
inserting
documents,
because
if
I
recall
correctly,
postgres
should
scale
fairly
well
unless
you're
trying
to
insert
millions
of
documents
per
second,
which
doesn't
seem
to
be
the
case
here,
but
I
I'd
like
to
hear
your
opinion
on
that.
So
that's
my
question.
For
now,.
A
B
Now
and
review
uses
mongodb
only
for
two
things.
The
first
one
is
to
store
nurse
requests.
I
mean
all
the
information
for
the
nurse
requests.
There
is
pre-process
before
by
hive
and
azure
data
factory
and
another.
The
second
thing
is
to
store
some
analytical
reports
for
how,
for
instance,
how
many
nurse
requests
has
been
open
and
closed
per
month,
and
things
like
that,
so
I
think
that's
it's
possible
to
use,
what's
grass
json
d
column,
to
store
to
store
everything
like
that.
B
A
D
Yeah,
that's
that
sounds
reasonable.
There
was
one
thing
that
I
wanted
to
mention,
but
of
course
I
didn't
write
it
down,
so
I
forgot
what
was
it?
Let
give
me
a
second,
maybe
I
can
recall
it.
A
Go
that
helps
me
understand
this
better
as
well.
We've
got
the
other
components,
so
there's
adf,
it
looks
like
we
have
a
good
solution
for
that.
With
a
google
google
dataflow,
we
can
use
the
built-in
features
from
postgres
for
document
storage
instead
of
and
you
we
sh.
I
think
the
other
two
are
hive
and.
A
And
you
know
what
the
needs
are
there
and
I
will
say:
is
that
actually
kafka
or
a
queueing
system
is
something
we
would
actually
like
to
have
as
part
of
the
gitlab
product
and
that
actually
may
be
worth
integrating
and
making
it
part
of
the
part
of
the
product
as
well,
because
we
need
it
often
in
gitlab,
part
of
the
product
uses
database
tables
to
implement
cues
and
that
works.
But
it's
sub-optimal
at
scale.
Surely
so
we
may.
A
If
we
had
that
available,
it
may
not
just
be
our
team
that
would
use
it,
but
other
teams
as
well,
but
we'll
see
and
again
for
the
for
milestone
one.
The
only
thing
we
need
to
replace
is
adf,
because
it's
not
it's
an
azure
thing,
which
is
a
gtp
thing
for
milestone
two,
where
we
actually
integrate
into
the
product.
We
can
replace
other
things.
So
you
know
that
that's
not
a
instrument
plan,
but
that's
at
least
the
direction.
It
looks
to
be
going.
A
The
taylor
you
didn't
get
to
vocalize
your
comment
just
yet.
C
Yeah,
so
if
y'all
haven't
mon,
has
a
lot
of
extensive
knowledge
in
this
area,
she's
got
a
lot
of
experience
with
gcp
with
data
flow
with
bigquery.
So
as
we're
looking
to
migrate
away
from
azure
to
gcp,
I
think
she
could
be
a
very
powerful
resource.
She
also
has
a
lot
of
experience,
training
ml
models
as
well,
so
once
we
actually
start
getting
into
the
guts
of
the
the
model,
she'll
absolutely
be
a
valuable
resource
for
us.
Her
background
is
very
extensive
in
various
types
of
machine
learning,
including
npl
or
nlp.
C
That's
also
my
background,
so
when
we
start
doing
any
processing
of
issues
or
labels
or
things
in
the
future
happy
to
help
and
explore
that
so
yeah
she's
based
in
australia.
So
time
zones
are
always
a
little
interesting
we're
all,
I
think,
spread
out
all
over
the
place
so
finding
times
for
people
is
going
to
be
difficult,
but
she
might
join
these
meetings
at
some
point
too.
A
If
you're
watching
the
recording,
it
sounds
like
my
mom,
which
is
not
what
I'm
saying,
although
the
cool
yeah
and
I
think
alexander,
alex
you're
planning
to
is
part
of
onboarding,
you
were
planning
to
do
a
coffee
chat
with
mom
already,
and
I
don't
know
me
how,
if
you
talked
with
mon
in
the
past,
but
if
you
haven't
she's
a
good
person
to
catch
up
with.
Thank
you
challenge
between
poland
and
australia.
It
doesn't
make
that
easy
yeah.
I.
D
A
Understood
cool,
that's
all
we
have
in
the
agenda
anything
else.
We
want
to
discuss
as
a.