►
Description
Luke Kim from Spice.ai takes us through Spice.ai's approach to helping organizations build the next generation of building Data, AI, and Web3 native apps.
Spice.ai: https://spice.ai/
B
A
Everyone
listening
remotely-
this
is
our
fifth
community
computer
data
working
group
community
call
and
we're
very
fortunate
to
have
luke
from
the
spice
ai
team,
as
well
as
derek,
and
today
we're
gonna
learn
a
lot
more
about
their
platform.
A
In
particular,
it's
gonna
be
a
bit
of
an
unusual
session
because
we
had
a
different
team
that
wanted
to
join
but
they're
based
in
asia,
pacific,
so
they're
gonna
have
a
second
call
later
today,
we'll
probably
end
up
slicing
these
together,
but
nevertheless
I'm
looking
forward
to
hearing
everything
about
spikes,
ai
and
so
luke
I'll
hand
it
over
to
you
and
I'll.
Let
you
take
it
from
there.
B
Cool
thanks,
wes
thanks
so
much
for
having
us
I'm
going
to
share
my
screen
here,
set
this
up
a
second.
B
You
guys
see
that
okay
looks
good
cool,
so
I
was
just
gonna
just
demo,
a
bunch
of
our
stuff,
and
then
I
looked
back
at
the
previous
presentations
and
no
one
was
autoforming
and
there
was
a
bunch
of
really
awesome
presentations,
high
quality
ones.
I
was
like,
I
guess
I'll
have
to
do
a
dirk,
so
here
we
go
we'll
go
through
it
pretty
fast,
though,
and
hopefully
spend
a
bunch
of
time
in
the
product.
So
so
who
are
we
loot,
cam,
founder
and
ceo
of
spice
ai?
B
And
before
this
I
worked
in
azure
built
a
team.
There
called
azure
incubations
built
a
bunch
of
projects.
You
might
have
heard
of
one
called
data
which
is
a
distributed
runtime
for
building
applications
and
then
also
worked
on
a
bunch
of
developer
tools
and
infrastructure
for
github
azure
and
microsoft
and
spy
ceo
we're
an
early
stage
startup
and
we're
actually
a
protocol
labs
portfolio
company
as
well,
and
our
mission
here
is
really
to
make
building
data
and
ai
driven
applications.
Easy.
B
B
Then
we
went
to
public
cloud
and
our
cloud
applications
and
we
really
start
to
evolve
into
much
more
sophisticated,
distributed,
applications,
highly
reliable,
scalable
and
so
forth,
and
then,
as
like
an
increment
on
that
decide
to
build
these
containerized
applications
and
even
more
sophisticated
applications.
More
reliable
and
this
term
kind
of
came
around
called
cloud
native
and
a
lot
of
these
applications
were
built
on
platforms
like
kubernetes,
and
our
thesis
at
spice.
B
B
We
believe
that
these
data,
ai
native
applications,
are
really
going
to
be
native
to
building
these
intelligent
applications.
B
You
could
argue
that
a
lot
of
them
are
intelligent,
but
maybe
not
in
the
definition
of
intelligent
that
we
use,
which
is
by
really
leveraging.
You
know:
ai
techniques
like
deep
learning
and
machine
learning
and
so
forth.
B
And
so,
if
you
think
about
how
you
might
build
web
free
versions
of
intelligent
applications,
think
about
things
like
quickbooks
or
xero
accounting
packages?
Well,
how
would
you
build
that
in
the
web
free
world
with
things
like
intelligent
transaction
categorization,
for
example,
or
some
of
the
the
biggest
businesses
today
in
web
2
youtube
netflix
amazon?
How
do
you
build
like
ai
driven
recommendations
so
say
if
openc
wants
to
do
nft
recommendation,
when
you
go
there,
how?
B
How
would
they
build
that,
or
even
just
things
that
we've
had
for
years
and
years
like
gmail
with
spam,
filtering
and
fraud
detection?
How
would
you
build
that
on
say,
wallet,
12
messaging
for
protocols
like
that
xmtp.
B
The
other
thing
when
you
get
into
web
free,
blockchain
data
that
I'm
sure
a
lot
of
you
already
know
is
web3
and
blockchain
data
is
painful
right.
So
you
have
to
build
and
operate
these
massive
blockchain
nodes
that
you
know
tens
of
terabytes
going
to
petabytes
with
you
know,
chains
like
solana,
and
you
have
to
build
all
and
operate
this
massive
big
data
infrastructure
and
ai
infrastructure
and
ml
structure,
and
you
have
to
understand.
B
Smart
contracts
calls
logs
and
events,
and
that's
just
like
to
get
the
basics
to
start
making
data
in
ai
driven
applications.
B
So
a
lot
of
work
right
so
and
if
you
look
out
there,
there's
really
no
good
solutions
that
help
you
with
this.
So
there's
really
great
analytics
solutions,
but
these
are
focused
on
analytics
so
doing
the
graph
and
so
forth
and
really
they're
not
designed
for
massive
scale.
Training.
Big
historical
data
sets
bulk
data,
apis
ml
friendly,
and
so
what
you
find
is
most
companies
try
to
build
themselves
but
often
spend
a
lot
of
their
engineering
time.
B
So
what
we're
building
spice
aei
for
is
something
specific,
not
necessarily
for
specific
analytics
or
dashboarding
or
dapps,
but
specific
for
specifically
for
bulk
data
applications
and
machine
learning.
And
so,
if
you're
thinking
about
an
application,
production
application
really
needs
to
be
very
high
performance.
B
And
if
you
think
about
machine
learning,
you
need
to
be
able
to
query
and
fetch
millions
of
records.
Tens
of
millions
of
records
at
once
to
do
training
to
do
aggregations
and
so
forth,
and
our
specific
audience
here
is
really
developers
and
data
science
scientists,
people
who
who
are
building
applications
and
want
it
to
be
easy
right,
so
three
lines
of
code
to
get
your
date
web
three
data
into
numpy
or
pandas.
B
And
so,
if
you,
if
you
think
about
the
the
grid
that
we
had
here
before
spice
ai's
focus,
is
really
on
applications.
Machine
learning-
and
it
does
query
sql
query,
but
it
also
really
supports
these
massive
historical
block
data
apis
and
we're
building
it
to
be
as
ml
friendly
as
we
can,
and
so
here's
kind
of
the
general
platform
we
support
ethereum
bitcoin
and
we
run
our
own
nodes.
B
We
index
all
that
data
and
we
have
more
chains
coming
soon,
polygon,
solar
and
so
forth,
and
the
idea
is
that
you
can
build
your
applications
on
top
of
the
platform
and
access
the
platform
over
these
really
high
performance,
low,
latency,
apache
arrow
apis,
which
is
this
high
performance,
commonly
data
format
in
memory
format,
and
it's
used
by
projects
like
spark
and
pandas,
and
so
you
can
actually
use
these
libraries
as
well
directly
on
the
platform
and
use
all
of
these
tools
that
you
would
already
used
today
in
the
in
the
ecosystem,
and
so
with
that
we'll
get
to
the
actual
demo
part
and
we'll
take
you
through
a
little
bit
of
the
platform.
B
Now
it
is
early
access.
We
started
building
this
in
january,
so
pretty
early,
and
so
we
can
choose
some
of
it,
but
we're
still
building
quite
a
lot
of
it.
So
so
let
me
just
bring
up
this
really
quickly.
B
This
is
our
home,
page,
spice.xyz
and
so
we'll
just
log
in
and
because
we're
targeting
developers
and
data
scientists
you
log
in
with
github,
and
that's
just
kind
of
the
canonical
way
that
people
in
this
space
share,
we'll
probably
add
you
know
other
developer
friendly
logins
later
on
when
we're
out
of
preview.
So
we
are
still
in
preview
here.
B
If
you
look
here,
it's
a
very
simple
api,
but
it's
an
application.
First,
api,
sorry
interface,
and
so
we're
just
gonna
go
into
this
application
here
and
we.
This
is
really
just
to
help.
You
explore
the
api,
the
actual
product
itself,
a
sort
of
explore
explore
the
data.
B
The
actual
product
itself
is
the
api,
and
so
like
other
products
in
the
space,
you
can
just
do
really
quick,
sql,
query
and
you'll
see
it's
like
super
fast
and
everything
we
do
is
aimed
at
being
high
performance,
because,
if
you're
building
a
production
data
driven
application,
you
need
it
to
be
high
performance,
and
so
essentially
you
can.
You
can
explore
the
the
data
sets
that
we
have.
We
have
a
little
bit
of
a
data
set
reference
here
and
you
can
go
through.
B
We
have
tokens
nfts,
d5
data
sets
and
if
you
look
here
in
our
doc
in
our
docs,
we
actually
have
a
bunch
of
example,
queries
that
can
go
through,
and
so
a
couple
of
those
will
actually
take.
So
here
is
this
one:
just
getting
the
latest
block
number
and
I
put
that
one
in
there,
and
so
you
see
the
blaze
block
number
just
pulled
up
eight
seven
eight
and
come
over
to
either
scan
here
and
refresh
that
you'd
say
eight
seven
eight.
B
So
you
can
see
real
time
data
that
we
have
in
here.
If
we
continue
on
over
to
this
panda
over
this
python
tab.
What
I
mentioned
before
is
the
goal
is
to
make
this
as
easy
for
developers
as
possible.
So
we
have
our
sdks
and
one
of
our
sdks.
B
Is
this
python
sdk
and
you
can
see
here
it's
three
lines
of
code,
including
your
api
key,
including
your
query,
and
then
you
have
your
data
in
pandas
and
what
we've
done
is
abstracted
away:
the
use
of
apache
arrow,
which
is
high
performance,
grpc
connection
to
the
service,
which
means
you
can
get
all
this
data
down
super
fast
and
just
using
a
couple
of
lines
of
code
and
you've
got
into
pandas
or
numpy
or
any
of
these
other
libraries
matt
plotley
without
doing
a
lot
of
extra
work,
and
you
still
get
all
these
benefits
of
this
very
high
performance
interface
and
we
also
have
a
node
sdk
2,
we're
going
to
be
building
other
ones
like
go
and
rust
and
so
forth.
B
B
Keggle
notebooks
up
here
I'll
pull
this
down
because
I
can't
see-
and
so
if
I
look
at
say
one
of
these
dex
liquidity
props
over
in
kaggle,
I
can
easily
do
the
same
thing
so
here
we're
just
pulling
down
the
sdk
and
again
just
a
couple
of
lines
of
code.
This
time
the
query
is
a
little
bit
longer,
but
essentially
it's
just
a
query,
and
now
I
can
use
things
like
matplotlib
with
my
data
and
you
know
continue
to
have
that
ongoing
and
refreshed
and
super
fast.
B
So
a
couple
other
examples
of
queries.
We
can
do
things
like
you
know:
average
transaction
fees,
gas
fees,
kind
of
all,
the
things
that
you
would
expect
to
be
able
to
do
from
a
data
platform
in
a
space
and
they
should
all
re
running
pretty
fast
right.
So
here's
some
some
transaction
fees,
nft
api,
nfts
and
tokens.
One
thing
that
we're
pretty
proud
of
is:
we
have
a
very
good
detection
for
tokens,
and
so
here
we
have
esc-1155
and
it's
all
automated.
B
So
if
a
new
token
comes
up
on
the
chain,
we'll
be
able
to
detect
it,
and
we
kind
of
give
you
the
token
standard
as
well.
In
our
token
apis,
but
again,
this
is
really
just
the
start.
We're
just
getting
started.
We're
gonna
be
working
together
with
the
community
to
build
a
whole
bunch
of
data
sets.
Here.
B
We
also
support
just
like
a
lot
of
the
other
data
providers,
direct
access
to
our
nodes.
So
if
you
want
to
go
and
use
our
json
rpc
nodes,
we
you
can
do
that
too,
and
we
have
other
kind
of
value-added
apis
like
prices
and
gas
fees
and
and
so
forth.
B
So
that's
the
the
basic
platform
and
our
goal
is
really
first
to
provide
the
data
in
a
high
performance
bulk
data
way.
So,
for
example,
with
this
api,
you
can
go
fetch.
You
know
10
million
rows
at
once,
and
because
it's
coming
down
this
high
performance,
apache
arrow
api.
It's
kind
of
this
long-lived
connection,
that's
very
efficient.
You
can
even
stream
it
down,
and
so
the
next
step
after
that
is
what
we'll
talk
about
in
within
the
context
of
this
group
with
us.
B
Then
how
do
you
actually
apply
compute
over
that
data
and
start
processing
and
doing
some
more
interesting
things
with
it?
So
let
me
come
back
to
the
presentation
here
come
through
here.
B
So
if
you
think
about
the
things
that
we
can
do
here,
we've
we
see
it
in
kind
of
two
categories:
one.
We
just
love
the
idea
about
doing
compute
over
data
and
especially
working
with
biocoin
and
ipfs
and
protocol
labs
projects,
and
so
we
expect
to
be
able
to
contribute
to
the
working
group,
but
also
projects
like
which
I
can
never
pronounce,
and
you
know
we're
going
to
go
over
to
the
data
summit
in
lisbon
and
give
a
talk
there.
B
Philippine
my
co-founders,
let
me
go
over
there,
but
also
we
actually
want
to
really
work
with
the
group
to
integrate
some
of
the
projects
into
the
platform.
So
you
can
imagine
that
we
would
host
our
nodes
into
spice
and
then
enable
you
to
do
computer
ipf's
data
there,
but
combine
that
with
that
queried,
web3
data.
So
imagine
that
you
have
a
job,
you're
querying
data
and
saying
fetching
all
of
the
nfts
for
the
last
last
year.
B
Then
you
have
the
nft
all
that
list
of
nfts
in
your
actual
job.
You
can
then
take
that
query.
The
ipfs
data
and
get
the
actual
nft
do
some
processing
on
the
actual
image
and
then
send
the
result
back
to
the
chain
or
somewhere
else
or
even
generate
data
sets
for
other
spice
users,
and
so
we
think
that's
a
really
great
application
of
the
the
project.
B
So
we
have
the
the
platform
right
now.
As
mentioned
it's
in
preview,
we
have
a
bunch
of
customers
on
it
all
doing
some
really
cool
stuff.
B
Everything
from
nft
marketing
analytics
to
you
know
obviously,
trading
and
financial
applications,
but
nft
authenticity,
services,
wallet
messaging,
and
the
other
thing
I
wanted
to
mention
is
we're
also
developing
this
open
source
project
to
help
build
ai
driven
applications
easier
for
for
developers
and
essentially
being
able
to
use
a
simple
api
to
access
some
training
and
inferencing,
and
we
will
wire
up
that
data
ongoing
data
for
you
from
the
platform.
So
you
can
say
here
is
a
query.
B
I
want
to
work
on
this
data,
bring
it
into
this
runtime
now.
Let
me
train
and
influence
on
that,
and
the
longer
term
goal
for
us
is
to
build
this
into
the
platform
as
well,
so
that
we
just
make
the
entire
experience
of
building
a
data
and
ai
driven
application.
Just
really
easy.
The
data's
there,
the
runtimes
the
frameworks
there
and
all
of
the
ecosystem
projects
are
there
as
well.
So
a
really
fast
overview
of
the
platform.
You
can
check
it
out
at
spy.xyz.
B
There
is
a
wait
list,
but
if
you
pm
me
I'll,
let
you
in
and
yeah
thank
you
so
much
for
the
time
today.
A
A
And,
and
could
you
say
when
you
think
about
the
types
of
ai
applications
that
will
be
built
and
interacting
with
things
like
spice
ai?
A
Do
you
think
of
it
as
the
mode
of
operation
as
sort
of
a
more
batch
style
request,
or
does
it
get
more
scenarios
where
the
applications
will
need
frequent
requests
to
spice
ai
or
maybe
not
real
time,
but
where's?
Where
did
on
the
spectrum?
Do
you
think
that
a
lot
of
the
the
demand
will
come
from
yeah.
B
B
I've
worked
in
big
data
platforms
for
for
a
long
time
and
there's
always
this
kind
of
notion
of
like
the
batch
pipelines,
obviously,
and
then
real-time
pipelines,
if
you
think
about
say
if
you
want
to
work
with
large
scale
like
like
hundreds
of
millions
of
nfts
or
if
you
want
to
do
like
a
big
training
job,
it
has
to
be
bashed
to
some
degree
right.
B
I
have
to
get
all
of
this
data
in
history
and
learn
from
it,
but
then,
if
you
think
about
or
how
to
actually
inference
on
that
model,
it
needs
to
be
real
time
because,
as
new
data
comes
in,
I
want
to
use
that
for
inferencing
on
that
model.
B
So
you
really
need
to
combine
both
of
those
models
if
you're
going
to
be
doing
some
type
of
real-time
intelligent
actions
in
the
world,
but
it
doesn't
necessarily
have
to
be
like
perfectly
real
time
and
real
time
means
different
things
to
different
people.
So
when
we
first
set
out,
we
were
working
with
a
bunch
of
financial
applications
and
we
said
like
do
you
need
real
time?
They
said
yeah
we
do
and
we
thought
that
meant
like
hs
hft
like
sub
second
real
time
right,
then
we
went
back
and
actually
asked
me.
B
What
do
I
actually
mean?
I
said
actually
like
if
you
get
us
data
within
a
day
like
we're,
not
even
that
sophisticated
like
our
trading
strategies
and
like
a
day
is
real
time
for
us.
So
real
time
means
different
things
to
different
people,
but
I
think
you'll
eventually
need
to
combine
both
techniques
to
to
really
build
out
these
ongoing,
continuous
data-driven
applications.
B
And
again
that
can
be
like
the
examples
I
gave
before
it
could
be.
I'm
gonna
take
a
whole
bunch
of
data
trainer
recommendation
model
on
how
to
like,
like
what
nft
is.
Is
a
cool
nft
to
look
at
or
buy
or
it
could
be.
You
know
similar
like
spam.
Section
we've
had
for
years.
I
have
to
you
know,
look
over
a
whole
bunch
of
content
figure
out
what
spam
and,
what's
not,
then
I'm
going
to
need
to
have
ongoing
data
to
keep
it
updated.
A
B
I'll
say
one
thing
so
in
terms
of
that
use
case.
B
So
why
do
you
guys
do
such
hard
things
like?
We
already
have
customers
who
have
looked
at
the
project
and
are
like
this
is
super
awesome,
but
it's
too
hard
for
us
to
use
or
not
necessarily
too
hard
but
like.
We
want
to
focus
on
our
business
logic
right
and
we
don't
necessarily
have
time
to
set
up
all
this
infrastructure,
and
so,
if
we
have
a
way
to
gain
access
to
the
benefits
of
that
project,
be
able
to
do
compute
over
some
of
this
ipfs
data.
That's
awesome
and
please
go.
B
You
know,
help
help
do
that
for
us
and
and
they're
already.
We
have
a
couple
of
customers
who
already
like
really
struggling
with
like
massive
egress
amounts
of
data,
and-
and
this
would
save
it
would
make
this
much
more
efficient
right
if
we
can
bring
that
compute
to
where
the
data
closer
to
where
the
data
lives
close
to,
where
those
results
are
from
the
from
the
queries
and
so
forth.
A
You
know
it's
a
really
interesting
point
because,
just
like
in
you
know
the
web
2
world,
there
should
be
stacks,
there
should
be
back-end
services
for
back-end
needs
and
there
should
be.
There
should
be
tableau
of
web3
data,
which
you
know
in
some
ways
you
guys
are
going
to
be
able
to
help
build
that
need.
I
I
totally
agree.
So
that's
a
good,
that's
a
good
point
and
it
needs
to
be
built
that
way,
especially
for
less
technical
business,
folks
to
get
value
from
it
exactly
yeah.
A
Yes,
yes,
we
appreciate
it.
Thank
you
so
much
and
we'll
get
this
on
youtube
here
shortly,
one
one
last,
I
guess
advertisement
for
the
rest
of
the
group
lisbon
summit
november,
2nd
through
3rd
luke's,
going
to
be
there,
so
we're
going
to
have
lots
of
folks.
Please
please
let
us
know
if
you're
able
to
join
and
we'll
have
a
little
bit
more
content,
we'll
add
on
here
for
this
afternoon,
but
luke.
Thank
you
so
much
for
joining.
That
was
tremendous.