►
Description
Turkcell is a leading converged telecommunication and technology services provider. Operating within Turkey and internationally, Turkcell currently serves close to 48 million customers with a wide range of communications and digital services offerings. AI has been integral to Turkcell’s strategic vision, covering areas such as computer vision, natural language processing and intelligent automation to help transform customer experiences, drive operational efficiencies and bring a greater diversity of consumer and business innovations to market faster.
In this keynote, Turkcell and Red Hat will provide an overview of Turkcell’s AI use cases, and how they went about building a flexible, secure cloud platform to accelerate AI development, and associated delivery of intelligent applications. Technologies such as Red Hat OpenShift, JupyterLab, NVIDIA GPUs helped provide the much needed execution agility for Turkcell’s AI strategy.
You will also hear lessons learned across people, process, and technology transformation to help operationalize Turkcell’s AI strategy, and how these lessons have helped enhance work for data scientists, software developers, DevOps, and IT operations teams.
A
Good
morning,
good
afternoon,
good
evening,
wherever
you
are,
this
is
abhinav
joshi.
I
am
a
senior
manager
in
the
openshift
group
at
red
hat
and
I
have
over
20
years
of
industry
experience
in
a
number
of
roles
on
both
the
customer
side,
as
well
as
on
the
vendor
side.
In
my
current
role
at
red
hat,
my
team
focuses
on
developing
and
evangelizing
the
value
of
red
hat
openshift
hybrid
cloud
platform
for
cloud-native
workloads
such
as
aiml
data,
analytics
databases,
programming
languages
such
as
java
python
and
so
on.
B
B
Talking
about
the
data
science
and
ai,
actually
trickster
has
been
working
with
the
data
since
it
started
its
operations,
we
have
been
using
analytical
models
and
the
predictive
models
for
last
two
decades
and
it's
for
core
operations.
Basically,
and,
as
you
can
imagine
any
kind
of
activity
around
our
customers,
we
can
we
use
utilize
our
models,
new
acquisition,
churn,
propensity
product
switch
risk,
score
calculations
and
so
on.
We
are
trying
to
make
sense
of
our
customer
journey
across
the
channels
by
data.
B
Also-
and
this
is
the
data
science
part
as
we
call
it
when
it
comes
to
industrial,
ai
site
store,
is
rather
new.
Actually,
three
years
ago,
we
established
our
core
ai
team,
which
was
intended
to
focus
on
the
industrial,
ai
solutions.
What
I
mean
by
the
industrial
ai
solutions,
we
actually
group
our
solutions
into
several
pillars.
One
main
part
of
the
pillar
is
the
computer
vision.
What
I
mean
with
the
computer
vision
is
actually
the
steady
and
streaming
images.
B
I
mean
the
bold
videos
and
the
pictures
and
the
photo
analysis
and
the
nlp
technology
is
another
important
pillar
for
us
with
the
nlp
technologies.
We
include
text
analytics
speech,
analytics
with
the
speech
to
text
and
detect
to
speech
models,
and
we
are
actually
very
powerful
in
turkish
language.
We
created
quite
powerful,
nip
engine
in
turkish
language
and
even
we
created
our
own
ai
voice
chatbots.
It
is
also
a
part
of
nlp
technologies.
B
We
created
a
chatbot,
even
a
platform,
so
for
the
creation
of
new
chat
bots,
we
have
smart
rpa
solutions,
we
are
active
on
the
autonomous
cars
and
the
in-vehicle
ai
applications,
and
also
the
health
industry
is
one
of
our
key
pillars
that
we
are
focusing
on
today.
I'm
gonna
talk
about
some
concrete,
tangible
examples
and
use
cases
that
we
apply
our
air
technologies
and
I'm
talk
about
how
I,
how
we
utilize
our
platforms
and
use
cases
here
it
comes.
B
This
is
digital
onboarding.
Digital
onboarding
is
one
of
major
use
cases
that
we
apply
our
ai
technology.
We
have
a
self-service
applications
called
do.
It
is
a
customer
self
subject
self
service
applications
and
it
enables
our
customer
to
perform
any
kind
of
subscription
activity
and
the
service
operations
related
with
their
life
cycle
and
with
their
their
subscriptions.
B
Also,
we
are
utilizing
our
self-types
applications
too,
to
allow
our
new
customers
to
start
their
subscription
subscription
without
involving
any
deal
any
moment
on
the
call
center
site
and
the
dealer
site
how
it
works.
Our
ai
automatically
detects
the
id
of
the
customer
and
it's
captured
by
the
camera,
and
it
captures
the
image
I
mean
it
captures
the
photo
on
the
id
and
it
captures
the
some
related
fields
on
the
id
like
name
and
surname.
B
Then
we
ask
our
customers
to
have
a
selfie
on
their
phone
and
we
match
this
selfie
with
the
identity
photo.
Also,
we
got
the
we
using
the
ocr.
We
got
the
name
and
the
surname
information,
and
then
we
asked
our
subscriber
to
confirm
their
subscriptions
saying
that
okay,
I
confirmed
the
subscriptions
and
they
were
analyzing
the
voice
and
we
confirmed
the
confirmation
of
subscriptions
which
kind
of
ai
services
that
we
are
using,
which
kind
of
ai
actually
use
cases
that
we
are
applying.
B
Computer
vision,
ocr
speech
to
text
model
and
the
nlp
technologies
are
using
for
that
specific
use
case.
Another
lovely
use
case
is
it's
also
the
our
ai
and
division
technologies
applied.
It's
a
special
campaign,
we
call
it
in
turkish
language,
gulem
saygrum
set,
it
means
in
english
smile
and
they
make
you
smile.
It's
a
special
campaign
in
that
campaign.
A
ai
our
ai
service,
detects
your
your
face,
detects
your
smile
rate,
your
smile
and
based
on
these
ratings.
B
B
B
B
The
essence
behind
this
technology
is
for
each
use
cases,
we
don't
develop
the
same
code
and
the
same
scenario
over
and
over
again
I
mean
this
is
a
cloud
native
application.
It
is
working
on
the
public
cloud
and
it
is
available
to
open
up
anywhere,
I
mean
in
the
world.
We
started
this
kind
of
service
in
turkey,
then
they're
using
the
openshift
that
our
cloud
native
platform,
we
open
this
capability
to
other
countries,
other
subsidiaries
of
truth.
B
B
Lifebox
service,
this
system
service
that
you
can
upload
some
documents,
some
videos,
some
photos
to
the
cloud,
it's
quite
similar
to
google
cloud
or
the
apple
cloud.
There
is
a
such
kind
of
a
service
of
two
sites
and
we
created
a
nice
feature
for
that
service.
This
feature
creates
your
photo
and
select
the
ones
which
can
possibly
get
higher
ratings
if
you
upload
them
on
the
internet
or
instagram.
B
Another
use
case
fraud,
detection
on
online
trainings
and
online
exams.
As
you
all
know,
our
life
quite
changed
in
that
update
on
the
on
on
the
pandemic
and
all
our
life
activities
already
changed
or
altered
and
digitized
actually,
and
we
thought
about
how
we
could
apply
our
ai
technology
on
the
trainings
and
on
the
exam.
B
And
if
we
found
out
like
this,
we
apply
our
ai
face
recognition,
technology,
anti-computer,
vision,
capabilities
and
system
detects
your
photo
takes
your
selfie
match
with
the
identification
and
during
the
trainings
or
the
exam
continues
to
check
your
availability
and
your
presence.
Whether
you
are
online
connected
and
doing
the
desirable
things
if
something
undesirable
detected.
B
We
are
trying
to
apply
our
ai
technology
and
air
capabilities
in
terms
of
any
kind
of
business
that
trips
that
is
active.
Actually,
I
should
talk
about
first,
that
the
s2
star
we
no
longer
defined
to
sell
ourselves
as
a
court
auto
company.
We
rather
start
to
express
ourselves
as
a
technology
at
the
digital
service
provider.
B
In
line
with
that
position
and
game
positioning,
we
have
several
digital
services
and
quality
resources.
We
have
our
own
tv
service,
we
have
our
own
music
service
online
magazine
service.
I
think
we
all
these
services
they
take,
they
contain
contents
and
we
are
trying
to
enrich
our
services
with
the
ai
capabilities.
B
One
example
is
the
nlp
and
detector
speech
capabilities.
As
I
said
at
the
beginning
of
the
presentation,
we
created
our
own
ai
voice
and
it
is
artificial,
but
the
printer
caused
a
real
human
voice.
I
mean
it's
not
like,
like
airport
announcements
and
our
ai
wars
can
vocalize
any
kind
of
written
message,
individual
format
and
the
into
conformance.
B
B
B
It
is
too
much
effort,
I
mean
in
terms
of
human
effort
and
in
terms
of
making
the
right
analyzing
in
order
to
help
them.
In
that
process,
we
created
an
ii
service
based
on
the
computer
vision
and
the
speech
technology,
our
ai
service,
analyzes,
all
the
interview,
video
shot
by
the
candidate,
and
it
automatically
creates
a
profile
of
applicants.
I
mean
the
the
emotional
status
choices
of
words
like
how
much
did
he
talk
about
the
technology?
How
much
we
talk
about
the
customer,
etc
and
the
base
on
this
interview?
Actually
the
evaluation
automatic
regulation.
B
Our
ai
service
creates
the
report
which
helps
hr
department
to
actually
evaluate
the
candidates
beautiful
recommendation
engine.
Actually
it
is
very
early
and
the
most
popular
application
of
the
ai.
As
I
said
in
the
previous
slide,
digital
services
are
key
part
of
our
strategy
and
our
customer
valuable
position
and
the
heart
of
actually
the
digital
services
is
content.
I
mean
you
have
to
make
your
customer
to
consume
your
digital
content
as
much
as
possible,
so.
B
Engines
are
key,
actually,
the
key
mechanism
to
to
make
your
customers
consume
more
content.
They
are
working
actually
a
couple
of
different
ways.
The
one
popular
way
is
it
profiles,
your
taste
and
it
also
profiles
the
content
and
it
makes
suggestion
based
on
this
match
and
it
matches
your
your
taste
and
the
content.
Another
popular
is
a
collaborative
filtering
suggests
the
content
which
the
similar
profiles
to
you
already
enjoy,
but
you
don't
touch
them.
B
Chatbots
everybody
knows
about
this.
We
also
have
several
active
chatbots
that
we
are
actively
using.
The
key
thing
is
about
these
two
to
two
things
I
can
say
about
them:
one
is
rather
than
creating
different,
different
different
chatbots.
Each
time
we
created
a
chatbot
platform
which
you
can
easily
create
and
configure
your
own
chat
box.
This
is
one
assessment.
The
another
thing
is.
B
B
So
once
we
actually
created
on
the
openshift
environment,
you
can
build
a
different
services
easily
based
on
this
technology,
and
the
second
thing
is
that
we
are
trying
to
make
our
chatbots
as
much
as
possible
on
a
proactive
way,
not
to
answer
your
questions
about
your
topics.
Also,
it
is
suggesting
you
some
new
contents
or
new
packages
on
the
new
tariffs
based
on
your
profile.
B
B
B
We
train
1000
employees
regarding
data
science
and
the
different
fields
of
the
analytics
and
artificial
intelligence.
Training
part
is
not
a
big
deal,
but
the
question
was
that
what
would
happen
after
all,
these
people
to
get
their
training?
They
needed
an
environment
to
publish
their
skills
to
apply
what
they
just
learned,
make
data
science
without
concerning
about
the
infrastructure,
environment,
data,
access
data,
preparation
and
the
necessary
software
libraries.
B
Therefore,
we
created
a
data
science
playground.
It's
a
great
data
democratization
example
throughout
the
organization.
It's
a
cloud-based
environment
that
contains
all
the
necessary
software
packets
libraries
for
any
data
scientist
that
we
need,
and
it's
also
integrated
to
all
our
internal
data
sources
and
the
environment
is
gpu
supported
and
the
gpu
resources
can
be
shared
automatically
based
on
the
load.
B
This
is
a
high
level
solution,
architecture,
the
it
is
working
under
the
openshift
311
and
the
jupiter
hub,
and
the
jupiter
notebook
is
supported.
Openshift
is
response
actually
responsible
for
the
running
jupiter
notebook
which
is
assigned
for
each
user
in
isolated
environment
and
the
python
they
are
using
mainly
python
in
jupiter
notebooks.
B
Active
directory
is
also
supported
for
the
identity
management.
Persistent
volume
integration
exists
to
retain
data
in
a
long
period
of
time,
and
also
we
are
supporting
the
system
with
the
git
plug-in
to
share
your
codes
and
versions
easily
all
the
resources
I
mean
it
can
be
ram,
cpu
or
gpu.
Resources
can
be
allocated
per
user
basis
and
the
gpu
versus
json
in
place,
and
we
started
all
our
analytical
models
into
that
environment.
B
B
Like
I
mean
starting
from
the
don't
want
to
lose,
we
don't
want
to
develop
and
deploy
all
codes
from
the
scratch
and
repeat
the
development
trainings
and
the
development
process
each
and
each
time
we
don't
want
to
deal
with
the
infrastructure
and
the
environment
necessities
over
and
over
again
for
each
use
cases.
We
want
to
create
a
microservice
based
service
architecture,
which
is
integra,
which
is
very
easy
to
integrate
and
to
to
manage.
B
We
want
to
dynamically,
manage
the
software
and
the
hardware
resources
and
the
capacity
we
want
to
globally
implement
the
identity
and
access
management.
We
want
to
integrate
all
our
internal
data
sources
to
the
system.
We
want
our
developers
and
the
data
scientists
to
not
deal
with
the
environment,
libraries,
softwares
and
the
hardwares
just
to
do
their
job,
and
if
we
want
to
create
ai
services,
which
are
necessarily
cloud
native
and
accessible
anywhere
and
accessible
anytime
at
the
anywhere
in
the
world,
so
we,
this
is
also
high
level
structure.
B
After
our
solution
for
the
ai
platform,
ai
cloud
native
platform,
as
you
can
see,
we
already
utilized
various
threaded
technologies.
I
have
listed
our
reasons
and
the
needs
why
we
did
this
in
the
previous
slide.
Actually,
our
solution
for
that,
our
the
primarily
finding
for
the
our
needs
was
the
communities
I
mean
we
already
decided
to
use.
B
As
I
mentioned
previously,
it
is
always
the
best
to
develop
cloud-native
platforms
containing
ai
services.
I
mean
which
can
be
easily
deployable,
reused,
flexible
and
scalable.
On
the
cloud
with
our
ai
platform,
our
goal
was
to
gather
different
product
families
under
one
roof
and
to
make
them
available
to
all
our
customers
in
terms
of
software
software
as
a
service.
By
doing
this,
we
aimed
both
to
share
our
work
in
the
field
of
artificial
intelligence
and
to
create
new
revenue
channels.
A
All
right
thanks
a
lot.
That's
a
lot
of
great
work
for
the
social
good
and
very
good
use
cases
that
you
have
for
the
ai
technology.
Now
before
I
talk
more
about
red
hat
openshift,
I
wanted
to
share
some
market
data
on
the
value
of
containers
for
improving
the
ai
workloads.
So
the
results
that
you
see
on
the
screen
are
from
a
recent
survey
conducted
globally
by
the
industry.
A
Can
you
go
to
the
next
slide
in
hunch?
The
red
hat
open
shift?
Is
the
industry
leading
containers
and
kubernetes
powered
habit
cloud
platform
to
help
accelerate
the
ai
life
cycle,
be
it
your
data
center
public
cloud
or
edge,
and
it
provides
a
lot
more
than
the
fundamental
value
that
you
get
with
containers
and
kubernetes
because
of
a
lot
of
capabilities
that
we
have
added
into
openshift.
So
what
what
openshift
does
is?
A
It
simplifies
the
deployment
scaling
and
the
lifecycle
management
of
the
containerized
aiml
tool
chain,
such
as
jupiter,
notebooks,
python,
tensorflow
or
even
commercial.
Let's
say
software
from
our
partners
like
h2o.ai,
starburst,
presto,
selden
and
so
on
by
being
able
to
automate
the
day
one
to
two
operations
with
these
tools
and
then
being
able
to
ensure
the
faster
time
to
value
now
the
integration
with
the
hardware
accelerators
such
as
the
nvidia
yeah.
Can
you
go
back
to
the
previous
okay,
okay
yeah.
A
The
integration
with
the
hardware
accelerators
such
as
nvidia
gpus,
is
to
ensure
that
the
modeling
and
the
inferencing
task
can
be
seen,
can
seamlessly
consume
a
gpu
from
from
openshift
and
then
like
openshift,
with
both
the
self-managed
or
a
public
cloud.
Hosted
option
provides
a
consistent
way
to
perform
the
day,
one
to
two
operations
like
regardless
of
the
location
where
your
developers
and
the
data
scientists
are
working
on,
and
then
they
get
the
consistency
and
the
portability
of
modeling
and
the
application
development
workflows.
A
With
the
data
gathering,
preparation,
modeling
and
the
deployment
and
then
finally
openshift.
It
helps
to
extend
the
value
of
devops
to
the
entire
machine,
learning
life
cycle
and
being
able
to
enable
the
mlaps
and
the
much
needed
collaboration
between
the
various
teams
in
your
ai
project,
and
this
helps
helps
ensure
that
the
machine
learning
models
can
be
easily
be
integrated
in
a
continuous
integration
and
the
continuous
deployment.
Continuous
delivery
of
your
intelligent,
ai
powered
applications
and
then,
finally,
so
openshift.
B
A
Financial
organizations
such
as
rbc
and
discover
financial
are
speeding
up
the
delivery
of
intelligent
applications
for
various
use
cases.
The
department
of
agriculture,
food
and
marine
are
speeding
up.
The
farmer
grant
applications,
so
there
are
like
number
of
use
cases
and
we
heard
from
inranch
as
well
on
the
customer.
Experience
use
cases
as
well
so
back
to
you
and
lance
to
give
the
closing
comments
and
the
lessons
that
you
learned
that
others
can
benefit
from.
Thank
you.
B
Okay
and
final
recommendations:
empty
takeaways,
actually
the
organizations
can
be
different
at
the
different
organizations.
You
know
they
have
their
own
doing
things
styles,
but
I
can
suggest
that
and
I
can
recommend
that
in
order
to
have
a
successful
result
installed
inside
of
the
organization
for
the
good
use
cases,
your
artificial
intelligence
data
science
and
the
infrastructure
teams
should
be
working
in
a
collaborating
environment.
I
mean
they
should
be
collaborating
together
for
the
successful
result.
B
It
is
not
the
job
of
just
one
team,
I
mean
you
can
have
a
very
powerful
ai
team,
but
without
including
the
infrastructure
teams
in
the
environment.
You
cannot
have
successful
results.
The
another
thing
again
and
again,
I
am
repeating
okay,
I
mean
for
some
use
cases.
It
is
good
to
create,
you
know,
tailor-made
services,
but
for
the
most
of
the
use
cases
it
is
best
to
create
reusable
services
rather
than
developing
one-time
solutions.
Over
and
over
again
so
I
mean
whatever
you
should
build.
B
Whatever
you
build,
you
should
build
as
a
cloud
native
and
also
you
need
a
containerized
architecture
in
order
to
have
the
best
scalability
in
order
to
have
the
best
maintenance
performance
and
the
efficiency
and
the
service
access.
The
openshift,
like
platforms,
are
quite
successful
for
this,
because
I
mean
this
is
a
world
of
area
of
the
open
source
and
for
the
open
source
environment.
They
also
need
a
good
support
from
the
successful
vendor
successful
partner.
B
So
I
mean
the
with
our
partnership
with
the
reddit
and
with
the
use
of
the
relative
utilization
of
the
eureka
technologies,
which
was
the
safe
way
for
us,
and
I
can
also
suggest
you
to
stick
with
such
kind
of
a
way
get
a
good
partner.
It
can
be,
of
course,
some.
This
is
that
we
are
talking
about
openshift.