►
From YouTube: How to Build a Cloud Native Image Recognition Solution - Rolando Carrasco & Akshai Parthasarathy
Description
Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe 2021 Virtual from May 4–7, 2021. Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
How to Build a Cloud Native Image Recognition Solution - Rolando Carrasco, SPS & Akshai Parthasarathy, Oracle
Fn Project is an open source project for serverless functions. We want to share how we used Fn, AutoML Natural Language Processing, and K8s to improve older Optical Character Recognition (OCR) technology. The resulting image recognition application has been extensively tested at a bank in Mexico. Unlike older OCR tech, the new application does not use pre-defined markers such as “name,” “phone number,” or “address” to extract information. Instead, it works by recognizing these fields using machine learning – even if the ML model has not been trained with a document/picture of that type. The audience will learn how to create and scale serverless functions and Kubernetes. In addition, we will demonstrate how to leverage AutoML to read documents/pictures from any source, classify and extract fields of interest using a pre-trained ML model, store information to a NoSQL database.
https://sched.co/ekBw
A
Thank
you.
Thank
you
for
joining
us.
I
am
brandon
carrasco.
I
am
here
with
akshay
from
oracle
and
we
are
going
to
talk
about
how
to
build
the
cloud
native
image,
recognition
solution
using
oracle
cloud
infrastructure
and
services.
I
am
from
mexico,
and
this
is
my
first
time
presenting
in
kubecon
what
about
you
actually.
A
Okay,
so
let's
get
started
with
our
own
introductions,
I'm
going
to
do
it
first
and
then
axia
is
doing
himself,
so
I
am
from
mexico,
as
I
already
mentioned.
I
started
my
own
company
like
when,
like
10
years
ago.
Sorry
it
is
the
name
sps.
I
used
to
work
for
oracle,
about
2002
or
something
and
then
2010.
I
just
quit
and
started
my
company.
I've
been
working
with
oracle
for
several
years
in
oracle
technology
I
mean,
and
that
is
why
I've
been
joining
a
couple
of
advocacy
programs,
oracle,
ground,
breaker
program,
an
oracle.
A
Ladies
director
and
my
background
is
with
distributed
systems.
That's
what
I've
been
doing
for
a
while,
since
the
old
days
of
enterprise,
architectural
integration
and
web
services
as
way,
and
now,
with
these
microservices
cloud
native
architectures
and
principles,
my
contacts
are
there
with
twitter.
My
twitter
handler
is
in
the
slide
also
my
linkedin,
so
you
can
contact
me
with
those
through
those
channels,
so
I
think
actually
you
can
move
forward.
B
B
I
have
about
12
years
of
experience,
first
with
on-prem
technologies
and
later
with
cloud
infrastructure
across
various
organizations-
dell,
amazon,
web
services,
netapp
and
startups
after
that.
So
I'm
very
excited
to
be
talking
to
you
all.
If
you
want
to
reach
out
to
me,
my
social
handles
are
provided
here.
B
So
let's
talk
a
little
bit
about
the
story
itself
right.
So
this
is
a
presentation
that
has
a
lot
of
practical
implications,
the
implicit
seeing
that
we
are
in
kubecon.
This
story
is
about
using
cloud
native
principles
for
this
particular
very
practical
problem,
and
we
want
to
share
that
experience
with
you
how
we
created
these
apis
on
top
of
cloud
native
technology.
B
Now
the
use
case
itself
is
about
the
use
of
scanned
documents,
ids
contracts
and
so
on.
So
traditionally
there
are
a
lot
of
organizations
think
about
loan
processing,
think
about
bank
banks
and
those
kinds
of
industries.
Anything
that
requires
a
processing
of
documents
is
usually
employed,
usually
employs.
You
know
people
to
manually
look
through
those
documents
populate
those
fields.
Maybe
things
like
name,
your
bank
account
number.
B
You
know
other
identification
data
manually
and
as
people
do
it
manually,
it
can
be
slow
and
not
to
mention
it
can
also
be
error
prone
and
it
could
honestly
be
a
waste
of
human
resources.
People
are,
you
know,
made
to
do
much
more
complicated
tasks.
So
that's
the
problem.
We
want
to
solve
using
cloud
native
and
ai
technology.
B
So
the
context
for
this
is
that
you
know
you
could
have
multiple
channels
in
which
these
documents
are
provided
right.
Maybe
they're
provided
directly,
maybe
they're
provided
to
a
bank
teller
in
person
by
the
customer,
maybe
they're
provided
by
web
or
mobile.
Somebody
just
you
know,
uses
the
website
or
their
mobile
app
to
send
the
documents
over
could
be
sent
by
email
as
well,
and
there
are
many
other
ways
right
and
all
of
these
documents
usually
have
to
sometimes
be
validated
against
some
kind
of
enterprise
system.
B
A
financial
institution
like
a
loan
processing
company
might
want
to
validate
validate
the
account
number
of
the
person,
the
name
the
address
and
so
on.
They
also
might
need
to
be
validated
against
some
kind
of
government
registry,
maybe
like.
If
you
provide
an
id,
you
want
to
make
sure
that
that
id
is
genuine,
it's
not
a
false
id.
B
The
problem
was
around
automotive
loans
and
credits,
and
the
loan
approval
is
actually
done
only
after
a
number
of
checks,
like
credit
check,
id
check
for
fake
ids
and
so
on,
and
there
are
15
types
of
documents.
You
know
that
need
to
be
reviewed
across
addresses
loan
payments,
credit
scores
and
many
other
things
right,
and
you
want
to
do
it
in
a
cloud
native
fashion.
A
We
work
with
this
customer
who
has
specific
challenges
and
they
were
looking
to
streamline
the
the
time
that
they
take
to
approve
or
to
reject
a
loan
credit
all
right
or
a
loan
or
a
credit
for
automobiles.
So
they
were
reviewing
already
these
images
and
these
documents,
but
sometimes
they
they
had
already
a
scan,
but
the
quality
of
the
document
was
not
good.
A
For
example,
sometimes
they
their
documents
were
scanned
in
t
format,
so
it
was
very
low
resolution
also
another
thing
that
they
had
or
another
problem
that
they
had
is
that
sometimes
the
the
image
was
not
in
the
correct
position.
Okay,
so
they
needed
to
rotate
it
in
order
to
get
the
information
and
sometimes
while
they
were
doing
the
rotation.
Some
of
the
elements
were
where
were
lost
so
this
process.
A
It
is
very
relevant
for
them
because
they
use
they
actually
use
the
documents
in
order
to
approve
or
or
to
reject
the
loan
evaluation
all
right,
so
they
use
those
documents
to
get
very
specific
and
valuable
information
for
them
in
order
to
say,
okay,
this
is
a
good
person
in
terms
of
their
score
and
whatever
he's
showing
to
us
and
we
are
going
to
approve
or
reject
the
loan
okay.
So
this
was
kind
of
the
scenario
tons
of
information
documents
and
so
on
and
with
that
trying
to
approve
or
reject
rejected.
A
So
so
so
the
customer
can
can
realize
if
the
information
that
is
being
retrieved
from
the
document
is
with
a
high
level
of
confidence
or
with
a
low
level
of
confidence.
So
every
time
that
we
increase
that
level
of
confidence,
then
the
customer
is
feeling
much
more
comfortable
in
order
to
move
forward
with
their
processes
all
right.
So
this
is
our
main
goal,
so
we
are
going
to
be
talking
about
that.
We
created
an
api
for
them
and
actually
an
api
that
we
offer
to
many
other
customers.
A
So
so
you
can
get
the
idea
about
many
customers
using
the
same
api
and
then
all
these
cloud
native
principles
are
coming
together
in
order
to
to
make
a
good
offering.
So
the
first
step
is
to
get
the
image
recognition
solution,
which
is
via
the
api
that
I
just
mentioned
in
the
previous
slide
and
which
is
supported
by
many
oracle
cloud
infrastructure
services
plus
a
couple
of
google
services,
all
right
that
we
are
about
to
talk
about
and
then
for
this
specific
customer.
A
We
also
incorporated
an
rpa
bot,
which
is
actually
reviewing
the
information
that
we
get
from
the
first
step
and
then
making
the
decisions
either
to
approve
or
to
reject
the
loan
all
right.
So
those
two
things
for
this
specific
customer
were
created
to
solve
the
the
problem.
But
for
us,
sps,
sps,
we
created
the
api
and
a
couple
of
things
inside
that
api
in
order
to
offer
it
to
many
other
customers
with
cloud
native
solutions,
all
right.
So
that's
that
was
the
consequence
of
dealing
with
this
project
and
creating
this
api.
B
Right,
so
you
know
we
have
a
number
of
services
from
oracle
for
cloud
native
and
devops.
You
know,
but,
as
I
talk
through
these
services,
the
ones
that
are
marked
with
the
check
mark
are
the
ones
we
use
either
services
or
projects
that
are
relevant
to
this
particular
use
case.
But
before
we
go
into
that,
one
thing
I
wanted
to
point
out
is
that
oracle
cloud
is
committed
to
its
openness
of
its
cloud
platform
right.
B
So
resource
manager,
for
example,
is
an
infrastructure
as
a
service
offering
that
is
compatible
with
open
source
terraform
api
gateway,
builds
off
of
open
standards
like
open
api,
3.0
and
ap
api
blueprints,
and
so
on.
Functions
is
based
off
of
an
open
source
project
called
fn
project
that
is
led
by
oracle.
Container
engine
is
100
open
source
based
off
of
kubernetes
upstream
kubernetes.
B
The
streaming
service
is
compatible
with
open
source
apache
kafka.
We
also
leverage
the
cloud
events
standard
for
our
event
service.
You
know
you
can
specify
targets
using
fluency
agents
for
the
logging
service
grafana
for
monitoring
as
well.
So
the
first
one
we
want
to
take
a
look
at
is
api
management
and
api
gateway.
B
B
You
know
what
I
mean
is
you
should
be
able
to
receive
api
calls.
You
know,
as
is
the
case
with
the
application
to
receive
api
calls
at
scale
and
also
receive
them
and
send
them
to
a
back-end
service
that
could
be
load.
Balancers,
compute,
kubernetes,
serverless
functions,
so
serverless
functions
is
what
we
use
in
this
particular
case,
and
you
also
want
to
provide
the
necessary
security
and
rate
limiting
for
these
apis.
Api
calls
too.
So
rolando
will
tell
us
a
bit
more
about
the
use
of
api
gateway.
A
Yeah
sure
so
there
are
two
main
things
or
two
relevant
things
that
we
can
mention
about:
the
usage
of
oracle
cloud
infrastructure,
api
gateway
for
this
regard.
For
this
specific
project,
which
is
the
first
one
is
the
customer,
was
looking
to
have
a
solution
where
they
need
didn't
need
to
use
elements
in
the
on-premise
environment,
even
within
the
cloud
environment
that
they
already
had.
They
wanted
to
have
something
like
can
be
used
as
a
service.
A
So
so
we
needed
to
to
have
a
a
place
to
deploy
these
services,
and
we
realized
that
oracle
had
this
api
gateway
that
we
can
be
using
in
order
to
expose
the
apis
that
we
delivered
to
to
our
customers.
So
that
was
one
thing,
and
second
thing
was
that
we
also
didn't
wanted
to
to
manage
any
servers
or
any
anything
related
with
infrastructure.
A
We
just
wanted
to
use
the
service
and
expose
the
api,
and
it
happens
that
this
gateway
is
just
like
that
it
is
a
serverless
offering
from
oracle
that
we
just
provisioned
it
and
then
just
use
it.
So
it
was.
It
was
very
straightforward
to
do
that
and
for
us
it
was
fulfilling
those
two
things
that
the
one
the
customer
didn't
want
to
provision
anything
at
their
end
and
second
thing
that
us
sps,
we
didn't
want
to
also
lose
time
in
order
to
provision
anything
related
with
this.
A
B
Yeah
oracle
functions
is
a
serverless
platform,
so
functions
can
integrate
with
oracle
cloud
infrastructure
services.
Of
course,
it's
a
part
of
oracle
cloud
infrastructure.
It
can
also
integrate
with
platform
services
and
the
sas
applications
from
oracle
right
functions
is
based
off
of
an
open
source
project
called
fn
project.
That
oracle
leads
and
you
know,
being
based
off
of
open
source.
It
allows
applications
to
be
easily
ported
over
from
other
cloud
and
on-premises
environments
rolando.
Could
you
tell
us
a
bit
more
about
the
use
case.
A
B
Great
so,
of
course,
we
used
the
oracle
container
engine
for
kubernetes,
which
is
also
called
ok
containers,
as
you
know,
easily
package
and
move
apps,
and
they
package
just
the
code
and
the
dependencies
without
the
operating
system.
So
they
have
advantages
of
being
able
to
be
spun
up
and
spun
down
quickly.
B
A
So
we
just
mentioned
oci
api
gateway.
That
was
the
first
step
for
us.
Then
we
used
functions
as
well
and
for
our
first
phase
of
delivering
this
api.
It
was
good
enough
to
have
that
because
we
were
able
to
fulfill
the
customer
requirements.
We
were
processing
their
their
documents
and
recognizing
the
the
information
within
those
documents,
but
we
already
working.
You
know
on
a
phase
two
for
this
project
and
for
our
offering
for
many
other
customers,
and
then
we
we
have
these
alternatives
and
there
is
where
okay
came
into
the
picture.
A
One
is
that
sometimes
customers
are
looking
to
deploy
this
solution
in
in
their
own
cloud
environments,
and
they
normally
are
using
a
kubernetes-based
cloud
environment
for
deploying
applications.
So
we
are
able
to
deploy
our
solution
on
top
of
of
kubernetes
and,
of
course,
on
top
of
okay.
So
that
is
one
thing.
We
just
realized
that
customers
are
looking
to
deploy
at
their
end.
Why?
Because
of
the
documents
are
sometimes
are
very
or
for
security
reasons,
that's
what
I'm
trying
to
say
and
they
want
to
have
everything
managed
at
their
end
all
right.
A
So
for
our
from
our
end,
it
is
pretty
easy
to
have
what
we
just
mentioned
in
terms
of
functions
and
oci,
api
gateway,
plus
okay,
because
we
can
mix
those
things
together
and
and
deliver
a
very
good
solution
for
our
customers.
So
so
those
are
the
two
things
that
we
are
looking
for.
Our
phase
number
two
actually.
B
The
reference
architecture
over
here
just
shows
an
example
of
how
you
can
use
object.
Storage,
you
upload
an
image
and
you
use
the
oracle
functions
to
get
the
metadata
and
store
it
back
into
object.
Storage,
along
with
the
image
rolando,
would
tell
us
a
bit
more
about
this
more
complicated
use
case.
A
Yes
and
this
subject
of
storage
was
very
straightforward
for
us
to
incorporate
it
because
we
are
archiving
and
we
are
first,
we
are
saving
documents
and
managing
them
through
the
object
storage.
So,
every
time
that
we
receive
a
document,
we
are
restoring
it
in
the
object.
Storage.
Why?
Because
for
later
purposes,
for
example
the
tracking
of
the
document
or
if
we
need
to
retrieve
the
document
to
for
any
other
type
of
analysis,
then
we
have
it
there
and
secondly,
because
we
can
archive
it,
and
this
is
just
out
of
the
box.
A
We
just
generate
the
rules
for
the
arc
archiving
and
the
information
is
being
archived
automatically.
So
those
two
reasons
where,
where
our
reasoning
in
order
to
use
it,
one
just
to
store
every
single
document
that
we
are
using
and
secondly
because
of
the
archiving
capabilities-
and
it
happens
that
between
functions
and
object,
storage,
it
is
pretty
simple
to
to
mix
them
so
technically
also
it
was
very
straightforward
to
do
it.
A
Okay.
So
now
before
we
go
to
the
demo,
we
just
wanted
to
highlight
how
we
are
connecting
the
dots
and
connecting
the
dots
means
that
we
have
the
use
case.
We
already
explained
to
you
the
use
case.
Then
we
explain
a
set
of
oracle
services
and
then
we
are
going
to
connect
what
we
had
as
a
use
case
and
as
a
need
and
how,
with
oracle
technology,
we
are
able
to
fulfill
that
need.
A
This
was
the
set
of
seven
steps
that
we
were
able
to
deploy
and
and
to
make
them
together
in
order
to
deliver
the
solution.
So
there
is
one
first
step
to
transform
the
documents,
then
how
to
upload
them.
Then
how
we
classify
that
document
if
it
is
an
id
or
if
it
is
a
balance
or
whatever
and
then
how
we
curate
the
document.
Remember
that
sometimes
images
are
rotated
or
it
has
some
elements
that
are
for,
for
example,
low
resolution.
A
Then
we
go
to
the
ego,
rule
apis
and
retrieve
all
the
information,
and
then
we
tag
every
single
element,
so
we
can
identify
it
properly.
We
also
have
this
machine
learnings
in
order
to
identify
the
documents
and
obviously
the
recognition
about
the
elements
from
within
the
document.
So
so
these
are
the
seven
steps
that
we
were
following
and
those
seven
steps
are
going
into
this
elements
from
oracle
or
components
from
oracle,
so
we're
going
to
just
highlight
two
or
three
of
them.
A
So,
for
example,
for
the
documents
upload,
we
are
using
again
oci
api
gateway,
which
is
the
first
contact
from
our
customers
with
us.
In
order
to
receive
the
the
document,
then
the
object,
storage,
where
we
are
restoring
the
information,
also,
the
classification,
curation
and
tagging.
We
are
using
the
article
functions.
We
have
a
specific
oracle
functions
that
are
doing
that
and
then
we
have
other
other
elements
that
we
already
mentioned,
like
google
for
the
ocr
and
also
for
the
machine
learning.
Okay.
A
A
So
whatever
we
just
mentioned
in
previous
slides
about
the
oci
api
gateway,
dfm
project,
oracle
functions,
google
and
the
object
storage
is,
it
is
right
here,
so
you
can
see
these
steps
that
we
are
taking
and
how
the
information
is
moving
and
it
is
hitting
every
single
product
that
we
just
mentioned
and
how
we
are
using
it
and
again,
every
single
layer
that
we
are
using
is
is
is,
is
a
serverless
offering
either
from
from
oracle.
So
so
it
was
pretty
pretty
easy
to
to
deploy
it
all
right.
A
So
this
is
the
architecture
that
we
use,
and
I
mean
this
is
very,
very
simple
diagram,
but
we
are
going
to
use
it.
Oh,
this
is
the
the
things
that
we
use
so
functions,
getting
related
with
the
object.
Storage,
as
we
mentioned,
the
oci
api
gateway,
presenting
those
functions
into
the
external
world
for
our
third
parties
and
being
the
first
layer
that
we
received
the
document,
and
we
are
also
supported
by
the
google
apis
in
order
to
retrieve
the
information
from
the
document
and
also
for
the
machine
learning
itself.
So
this
our
environment.
A
This
is
our
ecosystem.
If
you
will
okay
and
then
we
are
going
to
to
the
demonstration
in
order
to
show
you
whatever
we
just
mentioned
in
previous
slides,
all
right.
Okay,
so
let's
go
for
it,
okay,
so
this
is
the
web
application
that
we
built
here
in
sps.
This
is
a
work
from
plinio
arvisu
and
arturo
gonzalez
from
my
team,
all
credits
to
them
because
they
were
the
ones
who
who
created
this.
I
am
showcasing
it
now,
but
they
are
the
ones
who
created
this
from
from
scratch.
A
So
this
is
the
web
application.
So
this
is
basically
what
an
end
user
could
be
using,
but
behind
the
scenes
it
is
happening.
What
I
just
explained
so
the
api
gateway
functions,
the
google's
apis
object
stories
and
then
all-
and
not
what
I
just
mentioned
in
previous
slides
all
right.
So
we
are
going
to
do
two
things
we
are
going
to
scan
or
not
to
scan.
We
are
going
to
upload
two
documents,
one.
It
is
a
high
level
or
a
high
confidence
level.
A
Sorry
document
which
is
in
in
a
very
good
shape,
if
you
will
that
we
are
going
to
evaluate
the
score
for
that
document,
which
is
a
telephone
line,
balance
for
arturo
actually
and
secondly,
we
are
going
to
upload
balance,
but
for
the
internet
subscription
for
our
company,
sps
and
you're
you're
going
to
get
the
difference
from
those
two,
the
first
one,
which
is
this,
which
is
a
telmex
balance.
This
is
a
document
that
we
have
the
scan
for
that,
and
the
level
of
confidence
for
this
is
is
very
high.
A
Let's
just
keep
in
mind
something
that
if
you
are
just
thinking
well,
this
is
basically
they
are
uploading.
The
the
document
into
the
google
api
and
just
retrieving
whatever
google
is,
it
is
getting
back
to
them.
That's
not
the
case,
because
in
the
normal
process,
what
we
have
is
this
rastering
process
which
goes
from
left
to
the
right
and
from
up
and
down,
and
it
is
basically
taking
the
elements
of
the
document,
as
I
mentioned
from
left
to
right.
A
So
if
you
can
imagine,
for
example,
in
line
which
is
cll
for
video
ds8
and
then
sue
estado
de
cuenta
presser,
it
has
nothing
to
do
the
one
in
the
right.
The
left,
sorry
is
the
address.
Okay
and
the
one
of
the
right.
It
is
just
a
paragraph,
it
is
informative
things
or
it
is
just
informational
section
of
the
document
all
right.
So
if
we
go
just
left
to
right
and
get
the
information
from,
it
is
not
going
to
be
very
valuable.
A
A
The
first
section
is
this,
which
is
what
we
call
the
quality
gate.
So,
as
you
remember
in
the
sequence
diagram
that
I
show
in
previous
slides,
we
had
the
first
step
or
one
of
the
first
steps
was
to
get
through
the
quality
gate
and
the
quality
gate
is
returning
the
general
score
for
the
document
and
the
general
score
of
the
document.
A
We
are
trying
or
we
are
targeting
to
have
it
above
the
70,
so
it
is
0.7.
If
we
have
a
general
score
above
above
0.7,
then
it
is
a
document
that
we
trust
that
the
information
that
is
that
this
api
is
retrieving
is
accurate.
Okay,
if
we
get
a
score
below
that
point
seven,
then
you
can
definitely
use
the
information,
but
the
level
of
confidence
is
not
going
to
be
high,
and
the
second
thing
that
this
thing
is
recognizing
is
the
language
of
the
document.
A
In
this
case
this
is
spanish,
so
it
is,
it
is
getting
the
information
in
the
right
language
in
this
case
again
in
spanish,
and
then
we
have
whatever
the
ocr
returned
to
us.
Okay-
and
this
is
basically
all
the
information
that
it
was
able
to
retrieve.
It-
has
now
a
little
order
itself.
Now
we
have
a
structure
it
in
in
such
a
way
that
it
is
not
the
left
to
the
right
and
up
and
down
that
I
just
mentioned
for
the
rastering
process.
A
We
have
very
clear
vision
about
addresses,
names,
numbers
and
so
on,
but
if
you
take
a
look
into
the
lower
part
of
the
application,
which
is
the
nlp
detected
labels
now
we
are
not
just
returning
the
information
from
the
document,
but
we
are
able
to
classify
it.
So
we
can
see
that
this
has
an
ex
an
expiration
date,
which
is,
we
can
say
which
is
the
name
in
spanish.
We
have
nombre,
which
is
name
in
english,
and
it's
clearly
gonzales
romer
arturo.
We
have
the
address
for
for
for
arturo
and
direction.
A
It
is
the
name
or
that's
the
word
that
we
use
in
spanish
direction
address
in
english,
and
now
we
can
get
or
return
the
information,
but
in
such
a
way
that
it
is,
it
is
understandable
what
it
is.
We
are
not
just
returning
the
information
and
that's
it
and
okay
use
it.
No,
we
are
able
to
classify
it
and
give
you
a
level
of
confidence
in
this
case.
A
Pretty
much
the
name,
the
expiration
date
and
the
address
you
have
a
100
level
of
confidence
that
this
is
the
right
information,
so
you
you
can
manage
this
information
at
your
will.
So
this
is.
This
is
a
very
high
level
of
confidence,
actually
the
top
level
of
confidence,
but
in
the
other
hand,
if
we
upload
the
other
document,
which
is
internet
subscription,
balance
by
this
company
by
the
name
axtell-
and
this
is
for
our
own
company
services
and
process
solutions
or
sps.
A
This
is
probably
you
are
thinking.
Well,
this
is
a
good
document.
It
is
clear.
The
image
is
not
that
bad,
but
there
are
some
elements
that
are
not
going
to
allow
us
to
read
to
return
a
high
level
of
confidence.
One
of
those
things
is
that
the
size
of
the
font
it
is,
it
is
kind
of
small.
So,
as
you
can
see,
there
are
sections
of
the
document
that
we
see
that
the
size
of
the
font.
A
It
is
not
that
good,
it
is
actually
pretty
small,
so
it
is
going
to
be
difficult
to
to
retrieve
it
that
actually
google
itself
is
or
has
a
minimum
regarding
the
size
of
the
phone.
So
that's
one
thing.
The
second
thing
is
that
some
of
the
elements
that
we
can
get
from
the
document
are
elements
that
can
be
confused
in
what
way
that,
probably
instead
of
a
zero,
we
are
recognizing
it
as
a
no,
for
example.
A
So
there
are
elements
that
may
be
complicated
to
to
recognize,
and
the
third
thing
is,
for
example,
in
this
square
that
we
are
putting
here
in
the
screen.
It
is
our
longest
strings
of
information
that,
in
this
case,
represents
something
related
with
the
irs
and,
as
you
can
see,
those
strings
are
first
of
all,
they
are
kind
of
long
and
in
order
to
identify
what
they
are
is
going
to
be
difficult,
because,
as
you
can
see,
they
have
nonsense.
They
are
just
pretty
much
a
sweet
sequence
of
characters.
A
So
if
we
go
below
to
the
quality
gate
score,
then
we
may
see
that
the
this
document
in
a
specific,
has
a
very
low
score.
0.53,
so
the
quality
gets
says
no.
This
is
this
probably
needs
a
manual
revision.
That
does
not
mean
that
you,
you
will
not
be
able
to
use
information.
Actually,
the
information
is
here
in
the
section
of
the
ocr
structure
text.
We
can
see
the
information
by
itself
and
in
the
nlp
detected
labels.
We
we
see
that
this
level
of
score
is
not
that
high
as
the
previous
one.
A
Okay
it
it
actually
did
the
whole
process.
We
are
able
to
to
process
the
whole
document
and
retrieve
from
from
the
apis
the
value
or
the
level
of
confidence
and
and
the
labels,
and
also
the
we
are
able
to
group
the
information.
But
this
is
not
so
so
high
in
terms
of
the
score
and
just
to
finalize
the
explanation
of
the
demo.
A
As
I
was
mentioning
the
rastering
process,
that's
something
that
we
use
and
we
actually
program
in
oracle
functions,
because
we
know
that,
because
of
the
position
of
the
elements
in
the
document,
we
are
able
to
recognize
what
it
is
and
to
identify
that
that
is
that
the
name
is
written.
Actually,
in
the
name
section,
the
address
is
written
in
the
address
section
because
of
the
position
in
the
document.
So
if
we
make
some
some
mathematical
calculations,
we
are
able
to
to
reach
to
to
identify
and
group
the
information
that
we
get
from
the
documents.
A
But
it
is
basically
mathematical
calculation
because
at
the
end,
the
rastering
process,
it
is
at
the
same
time
giving
us
some
a
matrix,
a
dot
matrix.
So
we
are
able
to
to
get
information
doing
some
mathematical
calculations
all
right.
So
this
is
what
we
wanted
to
show
you.
We
showed
two
different
documents:
two
different
levels
of
confidence
score.
We
explain
how
those
different
elements
are
in
the
mix
and
with
this
probably
you
are,
you
are
having
a
different
idea
of
what
we
are,
what
we
are
doing
and
what
we
are
achieving.
A
Okay,
so
thank
you
for
being
with
us
in
the
demonstration.
This
is
the
list
of
things
that
we
learn
for
deploying
this.
We
have
benefits
for
the
business
and
also
technical
benefits,
the
benefits
for
the
business,
and
we
are,
we
are
conclu.
We
are
giving
you
these
conclusions,
because
cloud
native
is
about
warring.
For
the
business
needs,
okay
and
we
are
using
the
technology
in
order
to
deliver
a
solution
for
that.
A
So
it
is
very
relevant
for
us
to
highlight
the
the
benefits
for
the
business
and
what
we
learned
with
this-
and
the
first
thing
is
that
we
streamlined
the
credit
and
loans
approval
and
delivery.
So
it
was.
It
was
very
clear
how
this
reduced
the
amount
of
time
that
they
were
using
to
review
the
documents.
So
it
was
maybe
a
90
percent
of
the
time
was
reused.
A
Also,
we
are
allowing
customers
to
upload
information
directly.
They
don't
need
to
go
to
the
branches
or
have
an
in-person
meeting
or
to
deliver
the
documents
or
scan
those
documents.
They
can
do
it
through
different
channels,
the
institution
it
is
actually
now
presenting
apis
for
third
parties,
so
this
is
also
very
useful
for
them.
So
now
they
are
offering
these
services
for
third
parties.
A
So
now
they
can
have
external
organization
using
their
platform
in
order
to
to
review
documents
and
and
and
be
part
of
this
chain
of
activities,
and
also
for
in
terms
of
manual
intervention.
We
reduce
that,
because
now
we
don't
have
these
large
or
very
large
teams,
reviewing
those
documents
and
for
the
technical
side
of
the
house,
we
streamline
the
deployment.
Obviously
it
is
very
easy
to
deploy
this.
A
We
can
actually
do
it
at
the
customer
infrastructure
if
they
want
that,
we
are
also
having
constant
changes
and
we
have
the
ability,
with
cloud
native
in
order
to
incorporate
those
changes
and
also
in
terms
of
scalability
ecd.
We
have
this
this
way
to
to
scale
and
pretty
easily
with
with
all
these
oracle
cloud
infrastructure
components.
So
so
this
is
kind
of
our
summary.
Hopefully
you
found
that
interesting
and
we
are
just
finalizing
sharing
with
you
our
contacts
again,
our
social
network
content.
B
So
you
can
check
out
the
github
repositories,
the
katakada
catacoda
scenarios
from
orlando,
and
also
you
know.
We
have
some
product
web
pages
for
oracle
cloud
infrastructure
if
you're
so
interested
in
giving
a
look
or
even
trying
the
cloud.
So
please
feel
free
to
reach
us
on
twitter
on
and
linkedin
our
handles
are
provided
there
you're
now
ready
to
take
any
questions
you
may
have.