►
From YouTube: AI in Blockchain for the Mortgage Industry with Devin Daly, TRUE CRO- FMSIG Mortgage Subgroup Update
Description
See this June's update where we delve into the incredible ai and blockchain conversation within the mortgage industry! Featuring our special guest speaker, Devin Daly, CRO of TRUE. Plus, the update on the state of blockchain in the mortgage industry by James Hendrick offers valuable insights and the latest developments in this field. Marvin Bantugan, the co-chair of our mortgage industry subgroup, is the host of the update.
Timestamps:
00:00 Welcome and presentation
00:25 Community information
03:26 Status of Blockchain & AI in the mortgage industry
11:15 Exploring the power of AI in blockchain for the mortgage industry with Devin Daly
19:30 What is TRUE
20:46 The TRUE platform
37:27 Questions
57:59 July’s update announcement
Join the FMSIG Mortgage Subgroup Wiki: https://wiki.hyperledger.org/display/CMSIG/FMSIG+Mortgage+Industry+Subgroup
Subscribe to the mailing list to receive the latest updates: https://lists.hyperledger.org/g/financial-markets-sig
Follow the Hyperledger Financial Markets Special Interest Group LinkedIn page: https://www.linkedin.com/showcase/hyperledger-financial-markets-sig/
----
Hyperledger – Open Source Blockchain Technologies
Hyperledger is an open source community focused on developing a suite of stable frameworks, tools and libraries for enterprise-grade blockchain deployments.
It serves as a neutral home for various distributed ledger frameworks including Hyperledger Besu, Fabric, Sawtooth, Indy, as well as tools like Hyperledger Avalon, Cactus and libraries like Hyperledger Aries, Ursa. Learn more about Hyperledger projects: https://www.hyperledger.org/use
Discord: https://discord.gg/rVCn4j7UFd
Case Studies: https://www.hyperledger.org/learn/case-studies
Training & Certification: https://www.hyperledger.org/learn/training
Tutorials: https://www.hyperledger.org/use/tutorials
Webinars: https://www.hyperledger.org/learn/webinars
Events: https://www.hyperledger.org/events
Vendor Directory: https://www.hyperledger.org/use/vendor-directory
Subscribe to the Hyperledger Newsletter: https://www.hyperledger.org/newsletter
Follow-us on Twitter @Hyperledger
Learn about Hyperledger Membership: https://www.hyperledger.org/about/join
#Hyperledger #Blockchain
License
Creative Commons Attribution license (reuse allowed)
A
This
is
our
June
2023,
a
hyperledger
presentation
so
before
we
get
started,
I
want
to
thank
the
financial
markets,
special
interest
group
and
the
hyperledger
foundation
for
their
ongoing
support
and
making
this
meeting
possible.
I
I
think
we
have
a
really
good
agenda
for
today,
but
before
we
do
that,
let's
go
ahead
and
take
care
some.
What
I
call
house
cleaning.
A
As
always,
please
note
that
this
meeting
is
being
reported
and
it's
under
the
umbrella
of
the
hyperledger
foundation.
So
we
ask
that
everyone
abide
by
the
antitrust
policy
and
code
of
contract.
The
antitrust
policy
states
that
we
avoid
discussions
of
companies,
specific
pricing
products
and
projects.
We
don't
make
negative
remarks
about
other
companies
or
products,
and
the
code
of
conduct
means
that
we
treat
each
other
with
respect.
We
never
discriminate
and
we
communicate
constructively.
A
We
fully
support
hyperledgers
policy
of
openness,
equity
and
inclusion,
and
for
new
participants
and
I
think
we
will
have
quite
a
few
today.
We
welcome
you
and,
if
you'd
like,
please
introduce
yourself
in
the
chat
and
if
you
have
any
questions
or
any
specific
areas
of
Interest,
let
us
know
in
the
chat
we
want
to
make
this
as
interactive
as
possible.
A
Okay,
here's
our
agenda
for
today,
we've
already
gone
through
the
general
introduction
and
welcome
Anti-Trust
and
code
of
conduct.
We'll
speak
really
briefly
about
some
of
the
hyperledger
Community
Information
James
will
walk
us
through
an
update
blockchain
in
the
mortgage
industry
and
then
we'll
have
Devin
Daly.
The
chief
Revenue
officer
from
True
will
walk
us
through
true
what
they
are
and
their
use
of
AI
and
blockchain
and
Q
a.
A
We
always
covered
this
slide
in
each
meeting
and
that's
very
purposeful.
This
is
to
reinforce
that
we're
all
on
the
same
blockchain
and
now
ai
Journey.
We
may
be
at
different
points
along
that
path
for
today's
discussion
in
particular,
and
we're
all
still
learning
about
AI,
especially
with
the
Advent
of
chat
GPT.
A
A
I'm,
just
gonna
go
through
the
next
couple:
slides
really
quickly.
These
slides
are
intended
for
those
people
that
are
new
to
the
group
and
would
like
more
information.
So
this
slide
provides
a
link
to
some
of
the
different
resources
that
are
available
through
the
Linux
Foundation
hyperledger
foundation
and,
as
you
can
see,
second
from
the
bottom,
the
link
to
our
subgroup
Wiki
to
access
that
information,
you'll
need
an
lfid,
and
this
information
will
help
you
get
that
and
then.
Lastly,
here's
just
some
blockchain
training
I've
taken
this
training.
A
B
Marvin,
thank
you
very
much.
If
we
can
move
on
to
the
next
slide,
all
right
so
starting
out
with
Community
by
nascom,
so
this
article
provides
a
high
level
overview
of
blockchain
and
distributed
Ledger
technology,
as
well
as
the
basics
of
smart
contracts.
So
excuse
me
everybody.
You
know
like
Marvin
mentioned.
If
you
are
new
to
this
group,
this
is
a
great
article
to
start
with.
If
you're
not
familiar
with
these
Concepts,
they
do
give
a
really
good
quick
overview
to
help
get
you
grounded.
B
The
article
goes
on
to
discuss
how
blockchain
and
smart
contracts
can
improve
loan
management
through
speeds,
security
and
cost
efficiency.
Some
of
the
topics
we've
talked
about
previous
previously.
It
also
goes
into
data
science
and
how
it
can
promote
improve
loan
management
by
providing
lenders
with
valuable
insights
into
borrowers
and
their
behavior.
So
this
is
information
such
as
credit,
squirting
fraud
risk
modeling
customer
segmentation.
All
of
these
can
be
analyzed
and
addressed
through
blockchain.
B
Moving
on
to
the
next
article,
so
JP
Morgan
Chase
is
developing
a
chat
GPT
like
software.
You
know,
as
we
all
know,
Jamie
dimon
is
not
a
big
fan
of
crypto,
but
he
is
very
bullish
on
blockchain
Technologies
and
what
they
can
provide
the
financial
industry.
So
the
company
has
applied
for
a
trademark
product
called
Intex
gbt.
This
last
month
on
index,
GPT
is
touted
to
tap
cloud
computing
software
using
artificial
intelligence
for
analyzing
and
selecting
Securities
tailored
customers
needs.
The
technology
has
a
range
of
possibilities
within
the
financial
industry.
B
So
JPMorgan
they've
got
about
three
years
on
the
the
trademark
in
order
to
get
approval,
it's
uncertain
yet
exactly
how
JP
Morgan
will
deploy
it,
but
the
article
does
close
discussing
how
financial
advisors
have
long
feared
the
arrival
of
Technology
that's
going
to
be
good
enough
to
displace
their
role
in
markets.
So
far,
those
fears
have
largely
been
yet
to
materialize
and
we
don't
know
that
we'll
be
seeing
them.
B
The
third
article
we've
got
on
here
is
from
Black
Knight,
so
black
Knight's
been
evaluating
the
applications
of
artificial
intelligence
to
such
areas,
analytics
avms,
Mass,
customization
process
Improvement.
The
article
discusses
how
Ai
and
ML
and
the
mortgage
industry
started
over
a
decade
ago
really,
and
black
knight
is
currently
using
an
AI
driven
point
of
sale,
transforming
the
customer
experience
and
delivering
a
highly
personalized
assistance
when
and
where
customers
need
it.
The
automated
workflow
walks
buyers
through
the
next
step
provides
a
self-guided
process
for
the
customers.
B
Just
a
couple
weeks
ago,
MBA
education
hosted
an
event
sponsored
by
Adventist
there's
aventis,
Marvin
bantugen
was
the
moderator.
We
have
a
panelists
that
came
together
talking
about
leveraging
Ai,
blockchain
and
newer
Technologies.
In
today's
challenging
environment
we
had
Leah
Price
from
figure
Mark
D'angelo
PB
from
finlocker
myself.
B
We
also
talked
about
how
companies
in
the
mortgage
industry
are
implementing
these
Technologies,
as
well
as
at
the
recent
MBA
technology
conference.
Ai
was
a
big
topic
of
discussion
there.
You
know
one
of
the
big
focuses
that
groups
were
looking
at
was
the
risk
using
chat.
Gpt
proprietary
data
potentially
allows
us
chat
GPT
to
learn
from
that
data,
which
can
then
be
shared
with
others.
So
there's
definitely
a
concern
about
that.
It's
really
all
about
how
it
comes
to
deployment.
B
A
No
way
other
than
it
was
I
think
a
very
interesting
panel
discussion.
It
was
very
interactive.
Everyone
on
that
panel
was
a
extremely
knowledgeable
about
the
different
Technologies
and
I
think
what
was
most
interesting
was
everyone's
ability
to
talk
about
some
of
the
challenges
and
how
companies
can
take
a
lightweight
approach
using
pocs
using
small
test
cases,
a
way
of
making
some
of
these
new
technologies
a
bit
more
approachable.
B
Yeah,
fantastic
thanks
for
sharing
that
Marvin.
So
if
you
get
an
opportunity
and
you've
got
an
hour,
you
know
a
great
presentation
to
attend.
We
who
do
have
a
deck
that
went
with
that
presentation
too,
that
has
a
ton
of
information
in
the
appendix
as
well.
So
if
you
guys
are
in
need
of
that
at
all
feel
free
to
reach
out
Martin
next
slide,
and
then
you
know
last
a
monthly
reminder:
this
is
our
wiki
page.
B
Do
you
know
take
a
a
look
at
the
link
below
we'll
drop
it
into
the
chat
as
well?
In
fact,
I
think
Alma
just
dropped
it
in
there.
For
us,
you
know,
save
it
as
a
favorite,
the
lfid
that
you
need
to
set
up,
there's
instructions
in
the
upper
right
again,
it's
a
free
service
to
to
set
up
if
you're
interested
in
any
of
our
previous
presentations
over
there.
B
If
you're
looking
for
a
different
additional
information,
as
I've
mentioned,
we've
curated
in
the
last
year
and
a
half
well
over
200
articles
on
blockchain,
we've
started
to
add
some
AI
articles
in
there
as
well,
so
feel
free.
If
you
don't
find
anything
on
the
the
wiki
site
that
you're
looking
for
you
know,
be
a
pleasure
to
have
you
reach
out
and
we'll
share
what
we've
done.
Marvin
next
slide
and
I'll
pass
it
back
over
to
you.
A
Hey
thank
you
James
great
information.
As
usual
next
I
I'd
like
to
introduce
everyone
to
Devin
Daly
he's
the
chief
Revenue
officer
at
true
and
a
board
member
of
the
New
Jersey
MBA
he's
an
experienced
senior
executive
with
the
demonstrated
history
of
working
in
the
computer
software
industry,
he's
skilled
and
has
significant
expertise
in
banking,
Enterprise
software
sales,
e-commerce
and
Entrepreneurship.
He
was
also
part
of
a
panel
discussion
yesterday,
titled
how
AI
brings
trust
to
borrower
data.
A
C
C
Even
you
know,
I
spent
the
last
five
years
at
a
large
Independent,
Mortgage
Bank,
where
you
know
we
consider
robotic
process
Automation
and
it's
a
robotic
process.
Automation
is
a
wonderful
technology,
but
we've
got
to
do
something
about
our
data,
because
the
only
thing
that
RPA
is
going
to
do
for
us
is
help
us
make
bad
decisions
faster
if
we
don't
have
a
good
basis
in
data,
and
so
when
we
think
about
manufacturing,
you
know
really
good
data.
C
It's
about
leveraging
machine
learning
to
do
that.
It's
about
finding
documents
in
direct
source
data
and
being
able
to
cross.
Compare
it.
You
know
again
when
I
was
in
the
mortgage
industry,
not
as
a
not
as
a
solution
provider.
One
of
the
things
I
was
focused
on
was
continuous,
QC
and
I
was
focused
on
that,
because
it
just
seemed
that
we
spent
an
inordinate
amount
of
time
at
the
end
of
the
process,
trying
to
make
course
Corrections
and
then
we
would
incur
what
seemed
to
me
to
be
all
of
these.
C
Unnecessary
fees
might
be
increases
in
dwell.
Time
might
be
suspense
fees,
you
know
whatever
it
equated
to
it,
just
seemed
unnecessary
and
then
something
I
had
never
done
before.
I
had
an
opportunity
to
look
at
what
was
closing
and
my
God
that
I
get
an
appreciation
for
the
mess
that
is
post-closing
and
you
know
I
thought
I
was
going
to
be
able
to
go
in
and
pull
reports
and
categorize
and
codify
all
of
the
various
issues
across
all
of
the
10
or
12
investors
and
yeah.
C
And
actually,
at
the
same
time,
I
was
evaluating
replacing
our
current
OCR
vendor
and
there
were
a
number
of
reasons
for
that
and
it
mostly
because
you
know
the
the
state
of
the
current
OCR
technology
was
there's
it's
not
no
human.
In
the
loop.
There
are
a
lot
of
humans
in
the
loop
there's
a
lot
of
humans
in
the
loop
and
in
order
to
go
from
a
68
per
page
classification,
accuracy
to
a
90
plus
per
page
classification
accuracy
and
by
the
way,
automation.
Quality
data
is
95.
C
C
So
is
it
really
any
better
than
what
I'm
doing
now
and
and
my
thesis
was
I've
got
to
get
above
90
95
I've
got
to
have
that
automation,
quality
data
extracted
after
you
got
to
have
great
document
classification,
because
only
then
can
you
extract
data,
and
only
then
can
you
compare
it
across
the
document
set,
which
will
give
you
automation,
quality
data
and
what
I
found
was.
You
know
One
That's
the
basis,
because
on
top
of
that,
that's
where
you
layer
Solutions,
that's
where
you
put
your
income
analysis.
C
C
True
I
was
really
impressed
with
quite
frankly
what
I
said
to
them
was
what
you're
doing
is
so
great
and
so
far
Advanced
with
regard
to
the
rest
of
the
industry
and
your
competitors,
but
you
don't
have
any
solutions
on
top
of
it
and
a
year
later,
we're
launching
in
our
Solutions
as
you'll,
see
in
a
moment,
so
you
can
go
ahead
and
Advance
the
slide.
Marvin
I
think
the
important
thing
about
it
is
where
we
began
and
why
we're
so
good
at
what
we
do.
C
Our
founder
and
CEO
Ari
gross
is
a
PhD
in
computer
vision
and
has
been
doing
this
and
is
so
passionate
about
the
pursuit
of
Excellence.
You
know
when
we
get
to
99.5,
where
we're
at
for
one
of
our
clients
in
terms
of
per
page
classification,
accuracy
he's
the
first
one
to
say:
how
are
we
getting
to
99a
right,
there's
no
resting,
there's
no
resting
for
us,
and
so
that's
where
we
began.
C
You
know
we
began
coming
out
of
a
research
lab
that
was
led
by
Ari
and,
as
you
can
see,
we
have
a
couple
of
Fairly
large
clients
up
there
and
we've
pursued.
Since
then
you
know
a
real
focus
on
the
mortgage
industry,
and
so
today
we
recognize
850,
plus
document
classes
and
we
extract
about
you
know:
8
000,
Plus
data
fields
which
gives
us
you
know
when
you
look
toward
the
other
end
of
the
spectrum.
C
Here
when
we
talk
about
solutions,
that
was
the
basis,
and
some
of
you
may
have
seen
this
I
think
we
presented
this
as
digital
mortgage
before
I
was
part
of
the
company.
What
we
at
the
time
called
our
our
audit
tool
where
we
can
certify
across
documents
that
the
documents
are
consistent
and
then,
if
there
are
inconsistencies,
we
can
throw
those
up
to
be
manually,
attended
to
right,
and
then
we
can
take
those
documents,
the
authoritative
Source
documents
and
compare
them
to
data
in
the
loan
file.
So
we
can
do
a
really
thorough
audit.
C
Now,
when
you
think
about
that,
for
a
moment,
we
were
really
using
that
and
still
using
that
in
a
post,
close
QC
world.
Our
objective
in
our
strategy
going
forward
is
to
still
start
there,
because
I
think
one
of
the
tenants
that
we
have
to
begin
with
is
like
any
technology,
especially
any
new
technology.
We've
got
to
experiment.
We've
got
to
have
a
proof
of
concept,
as
Marvin
mentioned
once
we
prove
it.
That's
going
to
help
us
with
with
the
really
difficult
part
is,
and
the
difficult
part
is
not
technology.
C
It's
the
humans
that
use
it.
It's
the
organizational
change,
it's
the
process
change
and
if
you're
not
comfortable
or
you
question
the
technology
and
or
you
don't
have
internal
Champions
stop
right
so
begin
with
the
proof
of
concept
and
and
our
recommendation
because
it
can
be
so
powerful
is
start
with
post-closing
QC,
because
in
post-closing
QC
now
you
can
add
a
layer
of
Intelligence
coming
out
of
there
right
on
top
of
all
of
what
I
just
mentioned.
C
In
terms
of
that
data
analysis
and
data
validation
and
data
audit,
we
have
business
rules,
so
we
can
ensure
that
you
know
the
note
rate
is
the
same
from
the
authoritative
source
to
all
the
other
documents
and
when
it's
not,
we
throw
you
in
exception
right
and
we
do
that
via
business
rules.
Why
do
I
mention
that?
Because,
when
a
business
rule
fails
that
begins
our
quantification
of
where
the
issues
are
in
your
manufacturing
process?
C
So
now,
within
a
couple
of
months,
I
can
give
you
Insight
you've
never
had
before,
and
now
we
can
sit
with
lenders.
What
we
do
is
say:
listen
here
are
where
your
defects
are
right.
Here's
where
we
started.
Here's
the
proof
of
concept
and
you've
done
the
manual
review
and
you
trust
the
data.
So
now,
let's
shift
left
now,
let's
go
into
the
origination
process
and
catch
these
the
way
we
would,
let's
treat
it
like
a
manufacturing
process
right.
It
is
a
manufacturing
process.
C
Let's
shift
left
and
deploy
this
solution
to
catch
it
at
the
point
of
entry,
because
you
know
we
all
know
the
110
100
rule.
It
takes
a
dollar
to
collect
and
correct.
It
takes
ten
dollars
to
correct
it
and
a
hundred
dollars
when
you're
done.
So
why
spend
the
hundred
dollars
in
the
mortgage
industry
it's
more
like
ten
one
hundred
and
one
thousand
dollars
right.
C
It
solves
for
the
organizational
change
working
backwards.
Let
people
become
comfortable
with
the
technology
and
comfortable.
Let
them
have
assurances
that
they're
catching
this
in
a
way
and
at
a
speed
that
humans
never
could
and
we're
not
here
to
replace
humans
right.
That's
not
what
this
technology
about.
You
know
we
talked
about
earlier,
replacing
financial
advisors,
I,
don't
think
we're
going
to
replace
financial
advisors,
I,
don't
think
we're
going
to
replace
loan
officers,
but
what
we
are
going
to
do.
They
are
going
to
be
the
intelligent
operators
of
AI
right
Leah
Price.
C
We
talked
about
Leah
earlier
she
educated
me.
I
talked
for
a
while
on
this
and
Leah
said:
oh
yeah,
there's
a
term
for
that.
It's
called
prompt
engineer
and
those
are
the
people
with
the
domain
knowledge
that
ask
the
really
smart
questions
to
drive
the
artificial
intelligence
and
so
we're
not
replacing
humans,
we're
bringing
them
to
their
highest
and
best
use
based
on
their
knowledge.
You
know
no
longer,
are
we
staring
compare
folks,
no
longer
are
we
document
classification
or
data
entry
people
we're
using
our
higher
order
skills.
C
So
what
does
the
technology
do
and
a
true?
How
do
we
specifically
do
it?
So
OCR
technology
has
been
around
for
years
at
true.
We
have
our
own
OCR
engine
but,
most
importantly,
we've
deployed
machine
learning,
machine
learning
to
learn
documents,
but
machine
learning
also
to
learn
from
where
we
still
have
humans
in
the
loop.
You
know,
as
I
mentioned
before,
we
have
a
bunch
of
customers.
When
I
tested
the
platform
out
of
the
box,
it
was
96.7
accurate,
but
what
does
that
mean?
C
It
still
means
that
I've
got
to
touch
3.3
percent
of
the
documents.
Now
that's
vast
different
than
touching
100,
it's
an
even
you
know
and
a
vast
difference
from
touching.
You
know,
30
of
the
documents,
that's
a
90
reduction
in
fact,
but
how
do
we
do
it
right?
We
don't
use
OCR
technology
the
way
it
used
to
be.
We
use
contextual
classification,
which
means
you
know
for
for
for
a
guy.
You
know
not
a
guy
in
computer,
a
PhD
in
computer
vision.
That's
me,
it's
I,
explain
it
the
way
you
know.
C
Ari
explains
it
to
me.
It's
the
way,
a
human
reads:
the
document
we
have
a
large
Corpus
of
mortgage
related
terms.
We
know
what
documents
you
know
they
occur
on,
and
so,
when
we're
reading
those
documents
we
can
say
this
is
an
income
statement.
This
is
a
W-2.
This
is
a
closing
disclosure.
This
is
in
LA
right
and
we
classify
accurately
and
then
can
extract
the
data
right.
C
The
other
thing
that
we're
actually
seeking
a
a
patent
on
right
now
is
is
something
we
call
provable
correctness
and
the
basis
of
provable
correctness
is
for
one
thing:
you've
got
to
prove
that
you're,
more
accurate
than
a
human
is
doing
the
same
task
and
that
you
can
improve
over
time.
Humans
tend
to
you
know,
reach
a
ramp
of
performance,
there's
a
lack
of
domain
knowledge
to
domain-specific
knowledge
when
I
get
that
domain
specific
knowledge,
my
performance
doesn't
improve
with
machine
learning
and
with
machine
learning.
C
So
you
keep
your
learnings
and
we
we
have
our
our
Global
Learning
catalog
and
that's
the
way
we
assuage
those
concerns.
So
let
me
pause
here
for
a
moment.
Are
there
any
questions
about
how
we
do
it
specifically
and
obviously
I
can't
go
into
everything?
But
if
not,
then,
let's
roll
to
the
next
slide.
A
Devin
I
I
do
have
a
question
because
what
you
said
on
provable
correctness
really
stands
out
to
me
because
I
spoken
with
a
bunch
of
OCR
companies
and
every
time
they
come
in
and
make
their
pitch
for
one
of
our
clients
they
their
pitch
is
give
me
10,
000
samples,
two
months
of
time
and
a
hundred
thousand
dollars
in
fees
and
then
I'll
train.
My
software
for
your
problem,
I
mean
you
know
that
quote.
So
what
you're
talking
about
is
something
different
and
I.
C
Yeah
and
I
think
you
know,
I
think
what
makes
the
difference
is
a
true
we're
focused
on
the
mortgage
industry
yeah.
So
we
you're
not
going
to
incur
that
cost.
We
have
a
global
catalog
and
our
Global
catalog's,
probably
going
to
match
90.
If
not
95
of
your
needs
right,
you're
gonna
have
custom
docs.
We
need
to
recognize,
of
course
you
are,
but
the
vast
majority
of
the
docs.
You
know
we're
going
to
recognize
right
out
of
the
box.
C
So
there's
no,
no
learning
opportunity
that
we
need
to
do
yeah
unless
you
have
custom
docs,
and
we
do
that
rapidly.
We
can
actually
give
our
local
learning
engine
to
our
clients.
So
there's
no
Professional
Services
engagement
unless
you
say
which
a
lot
of
customers
say:
hey
I,
don't
really
have
somebody
that.
Does
that
and
you
know
we're
really
busy
closing
loans
over
here.
So
can
you
do
that?
C
But
we
actually
enable
our
clients,
through
the
local
learning,
to
actually
learn
the
documents
themselves
and
over
time
right
if
it's
their
humans
in
the
loop
right,
they're
folks
who
are
saying
hey,
you
know
we
have
a
threshold
like
any
other
AI
engine
or
a
confidence
level
if
it
doesn't
meet
our
confidence
level.
It
goes
to
a
human
for
review
when
that
human
re-categorizes.
It
we
learn
from
that
action,
so
we
you
know
we
take
that
into
account.
Next
time
we
make
a
decision.
C
So
we
had
a
little
overlay
here,
so
I
think
we
talked
about
this.
You
know
already.
We
have
a
mortgage
specific
AI.
We
are
BPO
free
right,
but
you
know
I
want
to
be
really
accurate
here,
I
think
one
of
the
Hallmarks
and
and
and
some
of
our
customers
and
Prospects
have
said
to
us
one
of
the
things
you
know.
That's
refreshing
talking
to
you
you're
not
telling
us,
where
you're
going
to
recognize
everything
and
you're
not
going
to
need
to
touch
anything.
That's
it's
just
not
true
we're,
not
we're
not
there.
C
Yet
with
the
technology
we
do
have
customers,
we
have
one
Mi
customer
who's,
99.5,
accurate
and
granted
they've
been
a
customer
for
three
years,
but
we
started
out
about
94
accurate
for
them
right
and
they
deal
with.
You
know
they
touch
a
lot
of
loans.
So
we
do
have.
We
do
have
BPO
Partners.
We
don't
have
our
own
BPO
shop
because
we
don't
need
it.
Quite
frankly,
so
100
BPO,
free,
100,
BPO
free,
is
going
to
get
you
about
95
of
the
way
there.
C
You
still
need
people
to
go
in
and
do
the
exception.
Processing,
hey
the
threshold
that
you
set
for
this
CD
was
75.
It
came
in
at
74,
somebody's
got
to
review
it.
Yes,
it's
a
CD,
so
we'll
learn
from
that.
The
next
time
we
classify
our
documents
we
talked
about.
We
have
really
robust
technology,
it's
not
just
OCR.
We
use
machine
learning
and,
and
again,
we've
invested
five
years
specifically
in
the
mortgage
space,
one
of
the
sexier
things
about
this
is
because
our
classification
is
so
good.
C
C
You
know
how
can
I
use
other
forms
of
AI
Predictive
Analytics
to
understand
what
customers
are
my
ideal
customer
profile
to
maybe
help
prioritize
for
my
loss,
what
prequals
they
should
be
following
up
with
right,
maybe
to
look
at
our
service
and
couple
this
with
our
servicing
data
and
and
develop
models.
You
know
to
understand
risk
in
the
portfolio,
so
the
next
challenge
I
think
for
customers
when
we
start
to
deploy.
This
is
how
can
I
monetize
all
of
this
additional
data
to
further
improve
my
operation.
You.
C
And
these
are
just
some
results.
I
do
want
to
point
out.
The
300
increase
in
under
underwriter
productivity
was
specific
to
Mi
99.5
we've
talked
about
one
of
our
Mi
clients.
There,
the
data
captured.
You
know,
8
500
data
elements
really
rich
sources.
The
other
thing
that
helps
us
do
those
8
500
data
elements
is
to
a
really
fine-grained
audit
across
the
documents
and
across
the
documents
to
the
data
in
the
loan
file,
and
we
do
that
really
rapidly.
C
C
Here
are
the
solutions
that
we're
working
on
right.
So
I
talked
to
you.
You
know
we're
going
to
launch
our
income
analysis
engine
next
month.
We've
got
our
post-closing
QC
solution
through
a
partnership
with
clear
Capital,
we'll
work
on
our
collateral
analysis
and
asset
analysis,
and
then
later
this
year,
we're
also
going
to
have
a
fraud,
detection.
I'll
say
it
again.
You
know
we're
really
fortunate
to
have
been
founded
by
Ari
gross
who's,
one
of
the
world
leading
experts
on
font
detection
right.
C
The
compliance,
Suite
compliance
or
true
quality
automation-
and
we
talked
about
our
audit
and
QC
here
so
I
think
what's
on
the
next
slide,
Marvin
is
is
kind
of
where
we
go
next
and
this
may
be.
You
know
what
you
know
where
we
started
kind
of
overall
discussion
about
Ai
and
blockchain.
If
that
makes
sense,.
A
Now
all
of
this
information
really
ties
together
well
from
an
OCR,
Ai
and
blockchain
perspective.
I
was
actually
in
an
MBA
technology,
Forum
meeting
about
three
or
four
months
ago,
where
they
said
that
the
IRS
had
actually
come
to
them
and
they
were
working
on
an
OCR
problem.
They
were
trying
to
trying
to
digitize
a
lot
of
the
tax
information
that
they
received
and
the
problem
that
they're
grappling
with
is
their
accuracy.
A
Threshold
was
about
65,
66
percent,
and
then
the
people
within
the
MBA
were
saying
that
from
an
industry
perspective
it
it's
more
90
95.
So
what
do
you
think
are?
Are
some
of
the
key
steps
that
a
company
or
an
entity
like
the
IRS
needs
to
go
through
in
order
to
get
their
accuracy
from
60
65,
initially
to
what
we're
seeing
at
as
an
industry.
C
C
C
I
think
it's
really
just
about
you
know
deploying
the
right
resources
I
mean
you
know.
Years
ago,
I
got
in
introduced
to
I'm
fortunate
to
live
fairly
close
to
Princeton,
University
and
I
got
introduced
to
several,
and
you
know,
if
you
know
anything
about
Applied
Mathematics
field.
Princeton
University
has
the
best
program
in
the
world,
and
I
got
introduced
to
12
phds,
who
were
focused
on
Applied
Mathematics,
and
these
are
the
guys
that
solve
problems.
C
You
know
if
the
phds
and
Applied
Mathematics
that
can
solve
those
problems
you
know
and
and
by
the
way
going
from
65
to
95
is
a
mammoth
task.
Yes,
going
from
believe
it
or
not,
going
from
95
to
98
is
an
even
it's
probably
an
order
of
magnitude
even
larger
than
the
65
to
95.,
so
they're,
probably
going
down
the
wrong
is
going
down
the
the
right
track.
I
should
say:
forgive
me
if
anything
from
the
IRS
is
listening.
C
A
Yeah
I'll
see
if
I
can
dig
up
their
information.
One
of
the
limitations
that
they
were
facing
is
that
they
could
not
use
any
type
of
BPO
Services,
because
what
the
people
in
the
group
thought
was.
Okay.
If
you're
trying
to
go
from
60
to
65
to
290,
then
you
will
need
not
necessarily
an
army
but
people
to
actually
support
that
step.
A
C
Yeah
and-
and
you
know
what
here's,
what
I'll
tell
you
about
that
we've
faced
that
problem
and
we
faced
that
problem
currently,
as
we
think
about
expansion
and
I'll
just
say:
maybe:
expansion
a
little
bit
north.
There
are
laws
stricter
than
those
in
the
U.S
with
regard
to
Privacy
Information
and
data
Can't,
Be
Moved,
outside
of
the
country.
So
what
true
has
developed,
because
we
have
this
large
Corpus
of
knowledge
about
words
and
phrases
used
on
us
Residential,
Mortgage
documents,
we've
started
this
automatic
redaction
process.
C
So
potentially
you
could
use
a
BPO
process
if
it's
100
automatically
redacted
and
then
you
have
to
determine
right
if
it
can't
be
100
automatically
redacted.
What's
the
nature
of
the
information
that
I'm
giving
out,
and
is
it
coupled
on
this
document
or
across
documents
with
any
other,
so
I
think
that's
a
that's
a
problem
that
can
be
solved
pretty
readily
I.
Think
we'll
have
that
solved
in
the
next
three
months.
A
Oh,
that's
great
that
that's
a
very
encouraging
to
hear-
and
you
mentioned
using
word
recognition
because
going
back
to
the
MBA
webinar,
one
of
the
things
that
we'd
mentioned
with
the
MBA
audience
is
from
a
Chachi
BT
perspective.
It's
built
on
I
think
100
176
billion
tokens,
so
those
are
letter.
Combination,
word
pairs,
so
in
the
word
recognition
approach.
That
true
has:
how
does
that
compare
and
contrast
to
the
Token
approach
that
chat
GPT
uses?
Are
they
compatible
or
is
it?
Why
misunderstand
the
whole
process.
C
You
know
the
choice
and
so
you're
100
right
in
in
drawing
an
analogy
to
it.
It's
just
that
the
you
know
the
body
of
knowledge.
The
language
model
is
a
lot
smaller
for
us
and,
quite
frankly,
that's
why
we
just
did
a
test
against
Google
and
AWS
with
regard
to
pay,
stubs
and
W-2s,
and
our
results
were
stellar,
and
it's
because
of
that
contextual
classification.
A
Okay,
great,
are
there
any
questions
from
the
audience?
I
I
have
a
ton
of
questions
since
I'm
really
excited
about
this
topic,
and
and
I
saw
the
presentation
that
you
guys
did
yesterday,
but
I
I
don't
want
to
monopolize
at
this
discussion
since
we
do
have
Devin
here.
So
if
there's
anyone
else
that
has
any
questions,
please
feel
free
to
type
it
into
the
chat
or
or
just
to
log
into
the
discussion.
A
Okay,
if
there
aren't
any
other
questions
from
the
rest
of
the
team,
I
I
did
want
to
get
your
thoughts
Devin
on
how
blockchain
ties
into
Ai
and
what
you
guys
are
doing
around
true
I
I
when
I
was
speaking
with
Ari
at
the
MBA
Tech
conference.
He
says
that
blockchain
can
be
a
basis
for
that
common
understanding
of
all
of
this
mortgage
information.
A
C
Yeah
I
I'll
say
it
a
little
bit
differently
and
I'll
I'll
draw
a
direct
analogy
to
to
what
my
objective
when
I
was
still,
you
know
on
the
lending
side
of
this
was
you
know:
I
talked
about
continuous
QC
when
you
think
about
continuous
QC
right.
If
I
can
take
my
continuous
QC
efforts
and
somehow
guarantee
and
certify
the
actions
that
I've
taken
right
and
create
a
fast
way
to
audit
it
earn
my
loans.
You
know
I
used
to
say
you
know,
maybe
flippantly
aren't
my
loans
worth
more.
C
You
know
on
the
on
the
secondary
market
right
and-
and
so
you
begin
to
ask
that
question
and
you
say
well
how
really
could
I
Engineer
that
and
I
think
I
think
the
answers
in
in
Ari's
answer
and
the
way
I
thought
about
it
was
and
the
way
I
think
about
blockchain.
You
know
I've
heard
a
lot
of,
and
you
and
I
talked
about
this
Marvin
I've
heard
a
lot
of
people
say
I'm
going
to
build
an
application
on
top
of
blockchain
and
and
to
me
that's
the
antithesis
of
blockchain
right.
C
C
We
do
intake
documents.
We
do
intake
of
direct
source
data
right,
maybe
I'm
using
form
free,
maybe
I'm,
using
plaid
or
I'm
getting
income
data
from
from
a
document.
We
do
a
fraud
check
on
that
document,
broad
check's,
fine
right.
We
hash
that
document.
We
take
that
hash
and
we
store
it
in
our
system
of
record,
but
we
also
put
it
on
a
blockchain
step
one.
We
extract
data
from
that
document.
We
take
that
data,
we
put
a
hash
on
it.
C
C
We
make
a
decision
based
on
those
calculations,
do
the
same
thing
and
so
at
the
end
of
the
process,
what
we
have
is
really
a
completely
auditable
and
certifiable
process
that
says:
I
got
this
document
or
this
data
from
this
place
at
this
time
the
borrower
uploaded
it
at
the
portal,
a
borrower
you
know,
signed
into
form,
free
and
got
the
data.
This.
However,
we
obtained
the
data.
We
certify
that
process,
we
store
it.
We
extract
data
from
it.
Maybe
it's
just
an
XML
fire
alert,
Json
file
or
we've
extracted
it
from
a
document.
D
C
Are
a
lot
of
costs
there
there's
a
QC
cost,
whether
your
QC
is
internal
or
your
QC
is
external.
When
you
begin
to
you
know,
use
tpr
firms,
there
can
be
a
significant
cost
there,
and
so
is
there
a
way
to
say
to
the
tpr
firm
I've
done
a
lot
of
the
work
for
you
or
this
third
party
has
done
a
lot
of
the
work
for
you?
Is
there
a
way
to
lower
that
overall
cost
not
only
lower
the
overall
cost,
but
make
this
a
lot
faster
right?
D
A
C
C
D
E
E
C
F
C
E
C
D
E
C
The
top
500
counties
are
responsible
for
93
of
the
mortgage
transactions,
so
maybe,
let's
just
start
with
500.,
but
a
lot
of
it's
around
title
and
I
think
you
know
the
appraisal
process.
I
know
the
gses
are
really
laser,
focused
on
on
the
appraisal
process
and
Appraisal,
Automation
and
and
firms
like
clear
Capital
that
I
mentioned
before
you
know
that
we're
going
to
partner
with
are
are
focused
on
Expediting
that
as
well.
C
I,
don't
know
if
I'm
gonna
get
in
trouble
for
saying
this,
but
there's
a
lot
of
people
with
their
fingers.
You
know
that
drive
the
cost
up
under
the
age
of
let's
protect
the
consumer,
so
there
are
a
lot
of
people
that,
in
the
middle
of
all
of
these
transactions,
particularly
at
the
back
end
and-
and
you
know
now
not
just
particularly
at
the
back
end
but
dirt.
You
know
during
the
process
as.
A
And
Jeffy,
if
I,
could
take
a
stab
at
answering
that
question
and
I'm
going
to
go
back
to
part
of
the
discussion
that
you
guys
had
Devin
yesterday,
that
a
significant
portion
of
the
origination
cost
is
due
to
compensating
the
sales
people.
So
my
personal
Theory
and
I'm,
probably
going
to
get
into
a
lot
of
trouble
with
this,
is
a
lot
of
what's
preventing
a
two-minute
or
even
more
realistically,
a
45-minute
loan
closing
process
is
the
social
inertia
that's
built
into
the
current
system.
A
So
we
built
that
application
and
that
was
up
and
running
and
still
is
up
and
running.
But
the
problems
associated
with
a
mortgage
are
several
magnitudes
more
complex
because
of
the
social
inertia
of
the
loan
brokers
of
the
different
parties
that
are
part
of
the
process
and,
honestly,
a
lot
of
the
risk
associated
with
it.
And
if
there's
any
type
of
fraud
losing
20
000
is
significantly
less
impactful
than
losing
500
or
average
cost
of
a
house
now
350
400
000..
B
B
B
The
reality
is
I,
see
the
process
being
shortened
down
to
two
and
three
days
over
the
next.
You
know
several
years,
but
I
don't
know
that
until
you
know
much
like
I
think
Devin
brought
up
until
you
get
all
these
title
companies
and
others,
you
know
integrated
into
a
you
know
similar
platform
or
connection
methodology.
B
A
Yeah
and
one
last
thing
that
I
want
to
add
in
this,
we
actually
built
a
proof
of
concept
for
a
loan
origination
process
where
we
could
get
through,
and
this
was
simplified
through
the
entire
loan
process
within
a
two-hour
window
using
Integrations
using
apis.
So
technically
it's
feasible.
It's
not
the
technology,
that's
the
problem,
it's
the
process
and
the
regulation
and
for
the
amounts
that
we're
talking
about.
Maybe
it's
a
good
thing
that
it's
slow
and
I
I'm
I'm,
really
impatient
I
in
I,
I'm,
a
tick,
tock
user.
Anything!
That's
three
minutes
long!
A
A
F
G
F
Sometimes
forget
that
this
is
a
a
risk-based
process.
Devin.
What
did
you
say?
You
guys
are
like
97
accuracy
on
those
docs.
Now?
Yes,
that's
awesome,
that's
awesome.
So
that
means
only
three
percent
of
the
time:
you're
gonna,
miss
and
Marvin's
right.
The
loan
is
about
350
000.
So
if
you
do
a
hundred
loans,
three
of
them
are
going
to
be
bad.
Now,
when
you
make
a
good
loan
as
a
lender,
you're
gonna
make
what
two
grand
well
in
today's
market.
F
The
profitability
is
actually
very
close
to
zero,
but
on
a
bad
loan,
you're
gonna
pay
back
the
entire
amount.
So
on
those
three
percent,
you've
now
lost
a
million
dollars.
How
many
loans
do
you
have
to
make
to
make
that
up?
You're
not
going
to
make
it
right
so
remember
this
is
about
this
is
about
risk
averse
business
people
trying
to
make
sure
they
don't
lose
money,
and
so
it's
never
going
to
be
a
two
minute
loan.
It
might
be
a
two-day
loan,
it's
more
likely
going
to
be
a
week.
One.
E
F
Is
what
happens
all
the
time?
That's
what
yeah!
That's
why?
What
Devin's
talking
about
today
is
so
exciting,
because
in
a
paper-based
world
we
have
errors
all
through
the
process
and,
as
Devin
pointed
out
the
longer
it
takes
to
solve
it,
the
more
expensive
it
becomes.
And
if
you
look
at
JD
Powers
surveys
of
borrowers,
the
thing
that
makes
them
the
angriest
is
being
getting
to
the
closing
table.
F
Thinking
it's
all
done,
and
it's
not
done
now
tread
the
changes
in
trade
change
that
a
lot,
because
now
you've
got
to
redisclose
everything
and
wait
another
three
days
and
that
punched
a
hole
in
that
problem,
but
it
gave
borrowers
another
reason
to
be
upset.
The
the
point
of
it
all
is
is
we're
not
lending
against
cars,
we're
lending
against
a.
B
F
Expensive
very
massive
asset,
and
it's
not
the
lender's
money
that
there's
giving
away
right.
It's
somebody
else's
money
and
it's
a
complex
system.
That's
never
going
to
be
like
a
car
loan,
but
with
software
like
what
we've
heard
about
today
from
Devin
and
with
other
blockchain
applications,
it'll
get
a
lot
better
right.
E
A
E
E
Have
to
do
this,
why
do
you
have
to
do
this?
How
come
you
can't
do
this
I,
don't
understand
this
younger
people,
they're
gonna,
be
like
it's
this
old
guy
doing
complaining
about
you
know.
You
know
what
a
blockchain
is.
Folks
I
went
into
my
Verizon
not
to
give
them
a
plug
that
a
new
phone
on
Verizon
about
a
year
ago
and
I
was
just
in
the
office
and
I
was
just
setting
the
phone
up.
You're
like
why
can't
I
pay
some
crypto
I,
don't
get
it
yeah
unless
you're
a
tech
company?
E
B
G
May
I
may
I
add
something
else,
because
Rick
was
saying
that
it's
a
a
risk
averse
industry
I
also
would
say
it's
a
lazy
industry
in
the
sense
that
usually
they,
unless
the
unless
Fanny
or
Freddy,
push
or
bring
up
bring
out
something
that
says
that
they
should
do
this.
They
tend
to
just
keep
on
doing
things,
the
same
way
and
they're
they're
not
so
prone
to
innovate.
G
So
only
if
they're
forced
to
move,
because
in
in
during
the
the
pandemic
during
covet,
they
were
they
had
to
update
how
they
were
closing
yeah.
They
could
have
done
that
way
before
because
the
technology
existed
way
before,
but
they
never
did
it.
They
didn't
do
it
until
they
were
forced
to
because
people
couldn't
go
to
sign.
G
So
if
they
couldn't
go
to
sign,
they
couldn't
close
business
and,
and
only
then
did
they
Implement
that
part
and
now
they're
working
with
it,
but
they
would
never
have
done
it
before,
because
no
one
was
forcing
them
to
do.
It
I
think
that
there's
also
a
bit
of
that
in
in
the
mix
right
that
it's
an
industry,
it's
not
the
regulations,
I
think
it's.
G
C
I,
this
is
gonna,
be
a
maybe
a
first
for
me,
I'm,
usually
not
very
sensitive,
Maria,
I
I.
Think
we're
of
like
minds
of
this
I'd
probably
use
a
different
word
or
a
different
phrase.
I'd
say
that
the
mortgage
industry
is
filled
with
fast
followers,
not
necessarily,
and
and
I
can't
I
can't
take
credit.
It
was
a
former
boss
of
mine
who
said
described
as
firm
as
as
a
fast
follower
and
I.
C
A
I
I
think
those
are
some
great
points
and
everyone
this.
This
was
a
great
discussion.
I
think
those
of
us
and
I
think
it's
almost
all
of
us
that
have
been
in
the
industry
for
a
while.
We
understand
that
that
challenges
of
working
within
the
mortgage
industry
and
that's
why
we're
here,
hopefully
that
we
can
help
change
the
industry
and
make
some
money
from
it.
So
I
I
want
to
get
back
to
you.
Devin.
Was
there
anything
else
you
you
wanted
to
cover
on
the
presentation
or
okay.
A
Okay,
I
I
want
to
go
ahead
and
because
we
just
have
a
couple
minutes,
left
I
want
to
talk
about
what
we're
going
to
try
and
cover
in
our
next
in
our
July
meeting
and
I'm.
Trying
to
get
to
the
right
slide.
Excuse
me
in
our
July
meeting
we're
going
to
have
a
another
company
come
in,
and
this
is
going
to
be
I'm
trying
to
share
the
screen
here.
A
So
we're
going
to
have
Excuse
me,
while
it's
loading,
Sanjay
Kumar
Nishant,
he's
a
chief
operating
officer
from
in
Kane
and
it
in
provides
a
blockchain
platform
using
a
hyperledger
fabric.
That
brings
the
different
parties
together
onto
a
blockchain,
a
single
blockchain
platform,
and
let
me
get
to
a
brief
description
really
quickly
here.
A
Okay,
so
in
tainted
the
blockchain
enabled
structured
Finance
platform
that
gets
lenders,
issuers,
investors
and
other
parties,
like
Services
trustees
and
rating
agencies,
onto
a
shared
data
platform
for
these
transactions
that
use
as
artificial
intelligence
and
blockchain
for
a
seamless
flow
of
data,
with
zero
reconciliation,
complete
transparency
and
provenance
of
asset
and
data.
So
you
know
what
that
that's
a
big
claim:
I've
spoken
with
Sunday
a
little
bit.