What was it like to switch from for-profit
to not-for-profit status?
JB: It was all about impact to us.
Earlier, I wasn’t helping as many people
as I wanted to. My career has transitioned
from embedded programming to
e-commerce to ad tech and desktop
visualization. The next step for me was
How are the challenges different in
JB: Technologically, we face more or
less the same challenges as any tech
startup. What is different are the key
performance indicator and the policies.
For example, instead of asking how we
can monetize something or maximize
revenue with a feature, we ask how
many more crisis counselors can we
support if we build something and
how many more lives can be saved by
pushing a feature out.
With your bachelor’s degree in computer
science, what role has formal education
played in your career?
JB: A CS background can give you a
great foundation and depth of knowledge
you wouldn’t have otherwise. Is it a
requirement? No. The requirement is
really a desire to solve problems and
break it down into logical groups. Some
training—self-taught or through a
university—is needed and depends on
how much time you want to spend. Some
of the best developers did not necessarily
go through a CS degree. A technical
background helps, but some of the best
developers I have met were music majors.
It matters how much time you put in—
through a school or through yourself.
Rahul R. Divekar is a Ph. D. student in the Department of
Computer Science at Rensselaer Polytechnic Institute,
Troy, N Y. His research focus is on the intersection of
computer and cognitive sciences, exploring areas of
group dynamics and emotive analysis in conversations
to enhance collective decision processes using AI. He
has a master’s degree in I T from Rensselaer Polytechnic
Institute and a bachelor’s degree in computer engineering
from the University of Mumbai, India.
Nidhi Rastogi is a Ph. D. candidate in the Computer Science
Department of Rensselaer Polytechnic Institute, Troy, N Y,
where she is leading innovation in anomaly detection in
large networks using graph analytics. She holds a master’s
degree in computer science from the University of Cincinnati
and has extensive work experience in networks at Verizon
Wireless and GE Global Research. She is also committed
to social good by using her skills in securing cyberspace,
net works, graph analytics, machine learning, and AI.
© 2017 ACM 1528-4972/17/03 $15.00
identifiable information. The data is never
allowed to be downloaded and can be
accessed only through our API. We also
do background checks on all counselors
who join the team. On the technology
side, we use role-based authorization and
two-factor authentication for anything
that touches the data. Access to certain
systems is restricted, even to the in-house deputy. All the data sits behind
VPNs and firewalls. Most important, we
encrypt all data.
What data can crisis counselors see
concerning texters to be able to help
JB: We know it would be helpful if
the counselor had some background
information about the texter, but that
would violate texter privacy. Counselors
know only what texters tell them. We
do envision that we would want to use
machine learning to identify certain
patterns in texts to help guide the
What are you able to predict from the
data you have?
JB: For now, we focus on identifying
trends, given a particular time, place, and
words. For example, a working individual
calling in at 9 PM on a Sunday night may
be feeling suicidal.
What other technological challenges do
JB: We mostly face what other tech
startups face. Keeping the service up and
scaling well. We do want to use machine
learning to use the data in the most
efficient way possible.
How has technology kept up with
scalability and the growth of your
JB: Our hardware and software are
horizontally scalable, meaning we just
need to add one more server to the load
balancer to scale up. We have had to make
some changes from when Crisis TextLine
started out but nothing more than a few
tweaks here and there and some query
How does the architecture handle usage
spikes at the time of a large-scale crisis?
JB: During the 2016 U.S. elections,
we saw a spike. Our key performance
indicator is to respond to every text
in under five minutes. The health of
the queue is determined by how many
people are responded to in under
five minutes. Most helplines handle
their queue in the order of arrival.
Our proprietary system, based on
choosing certain words in the text sent,
determines the severity of the immediate
crisis and will put certain messages
higher in the queue than others.
If you expand Crisis TextLine
internationally, how will the
JB: We want to help as many people
as possible in every country. Perhaps
we would look at a multi-tenant solution
and bespoke software. There are a bunch
of local requirements to adhere to; for
example, the data might possibly have to
be stored in each country.
but some of the best
developers I have
met were music
majors. It matters
how much time
you put in—through
a school or