Before we start this episode of the Data Skeptic Podcast, I wanted to leave a quick reference at the front end just in case anyone might later be coming back here looking for some contact information. This episode will discuss a service called the Crisis Text Line, which is a 24/7 SMS service connecting people in moments of crisis with live trained counselors. If you or anyone you know might be in need of such a service, you can reach them by texting the short code 741741. The Data Skeptic Podcast is a weekly show featuring conversations about skepticism, critical thinking, and data science. Welcome to another episode of the Data Skeptic Podcast. I'm joined this week by Noelle Saldana. How are you doing, Noelle? I'm good. How are you? I'm doing really well. Thanks for joining me. So in this interview, we're going to discuss some work you did with the Crisis Text Line, but I thought to get started we could talk about how you got into data science. Sure, my background is a little interesting. So I am an applied mathematician by training and yeah, so I got my first job in the industry at Fox Interactive Media, which at the time owned my space. So back then, if you remember, it was the largest social network in its heyday. So my division dealt specifically with the ads that were running. So it was like the ad network, not just on my space, but the other Fox properties. And so that division eventually got spun out a few years later into an ad network called Fox Audience Network. And some who couldn't years that became part of the Rubicon project, but that's ultimately where I got my start in a group called the Solutions Group. If you can believe it at the time, back then we still had a petabyte of data. We had billions of ads that were coming in. And we had a huge Hadoop cluster and a huge green plum distributed database at the time. And that was really where my start was. And before data science was really cut on as a thing, we were already doing analytics at scale. So we were doing some really cool analytics, not only on the ad data, but the profile data and a lot of interesting digital media stuff, then had a stint at e-harmony, not doing all the matching, which most people would actually doing a lot more on the marketing side. So trying to understand how they optimize their campaigns and how do they outreach to all of their customers and retain them for the ones who didn't end up happily ever after yet. No, it was fun because it was one of those companies where ultimately they were happy when people left because they actually used the service and never needed it yet. And ended up at a green plum, which then had morphed into pivotal. And I've been there for the past four and a half years working with various clients across the board, understanding their digital data and how they can really improve their practices with analytics at scale. That sort of brings me to one of the, in the range of customers that I've dealt with was, you know, crisis text line. Great. Yeah. So let's get into that. It sounds like you had support from pivotal, from data kind and from crisis text line. So how did that whole project come together? Right. So one of the things that our leader, like Monica Jimenez of the Data Science Group had realized is that not only is there a thirst and a need, as you know, for data scientists in the commercial space, but nonprofit organizations are also in need of that sort of resource. And that's why we have great organizations like DataKine that are out there. And that if we, that there's ways that you can do small things, right, on your own, like participate in a data dive or do something sort of on the side, like through data, core through data kind. But as a company, they wanted to be able to say, how can we also give back? How can we give back with data science? And so they came up with this program that had initially announced it about, I think a little over a year ago now, to say, you know, what we're going to be able to do is donate data scientists time and not just a weekend or a day or, you know, 20% time, but to allow the data scientists to spend, to donate three months of time to an organization of their choice, you know, as a data scientist and give those much needed skills in a way that will have an impact on society, right? And it also gives the data scientists who have been more or more tenured in the company, you know, who've been there for three years, a little bit of a break, right? So while there's no shame in doing what we do for the commercial space, it's, you know, more emotionally fulfilling to be able to say that you have the opportunity to use those skills and do something for data philanthropy at the same time. Yeah, I think that's fantastic. And in fact, one of the reasons I was really excited to have you on the show was to share a story about an opportunity to do data science in an area that wasn't necessarily revenue driven. Not that there's obviously any problem with that, but applying the data science skill set to more charitable endeavors is something I hope we see an increasing amount of in our community. So maybe you could tell me a bit more about data kind, which is one of the organizations helping to promote that sort of activity. So data kind actually did the announcement with us when we announced Pivotal for Good, right? So for Pivotal, we have dozens of data scientists who, you know, in rotation will have, you know, eligibility to participate with this program. But we don't have any experience with nonprofit organizations. We have all of this experience qualifying leads and such, but not for the nonprofit space. And that's really where data kind comes in. They're a little bit of a matchmaker, right? All that glue that goes into running a data science project, like consulting gig, essentially, is something that they take care of for the nonprofit space. So they do the matchmaking between the nonprofits to see, you know, who's actually in a place where they are ready for data scientists, they have interesting data, they have interesting problems, you know, and qualify it in a way that they can then unleash that problem to their data scientists and then get the, make sure that the data scientists are set up for success. They have things, as I said, sort of short term projects like data dives where you go and it's a hackathon type thing, but also the program called data core, which is sort of, which is more long term, right? So they can say over the course of six months, which allows you to really have time to sink your teeth into a problem. And they'll help set up with the project management and the tracking and the kind of client liaison piece, as well as the backend piece, first helping set up the platform and the data. As I said, all that stuff is that is really unsexy when it comes to projects, but is absolutely key and necessary to make sure things run smoothly. They were the ones who helped us launch Pivotal for Good. And it's been a tremendous experience. I've, you know, done work with them in the past, but really morphing what they do with data core and to an opportunity for Pivotal for Good has been really, has been tremendous experience. Yeah, it sounds like a really great partnership. So I may be showing my age a bit with this next question, but as I grew up, I was sort of peripherally aware that there were, you know, hotlines you could call and moments of crisis and things like that. But, you know, as I've gotten older, it never really occurred to me that the younger generation, you know, teenagers especially are, you know, more kind of texting rather than calling. So I'm glad to see that an organization in this case, Crisis Textline, is there to fill that potential gap. So is that more or less their mission to sit alongside the traditional phone hotline? Right. Yeah, no, I agree with you. Well, I, you know, I come from showing my age. Like, we had beepers and only some of us did. I didn't even went in high school. So, you know, it's new to me to be able to actually be connected as a teenager to like a cell phone or this sort of a thing. Yeah, so Crisis Textline is the first 24/7 national issue-agnostic hotline that's text-specific. So I know that there are a few others out there that do use text, but I think in the past, they've been limited either to local regions. They're only for certain times of the day, or very, very specific issues. So for Crisis Textline, their goal is not only, you know, they don't want to just help out locally, but they want to really, really help at scale on a national level, which will ultimately have like a much deeper impact on society. Oh, it's really excellent. So how did Crisis Textline get off the ground? So the way that it started was that Do Something had actually sent out a text message about volunteering and got an unexpected response back from it. So that was a text message that said, he won't stop raping me. He told me not to tell anyone, "Are you there? It's my dad." And that message is what sparked the CEO do something, Nancy Lublin, to found Crisis Textline so that we could do something about this, so that there's a place where all of those messages can eventually go, and somebody could actually talk to someone who will then be able to help them. Can you give me a little background on what Do Something is? Do Something is a national organization aimed at youth, so from the ages of teenagers to their mid 20s, to do more volunteer things, to do something in their communities. Oh, that's really great. But we should take a step back before I kind of interrupted you. So Do Something sent out this SMS campaign message and got back something they weren't expecting, you know, someone definitely in a moment of crisis who needed some help. How did that inspire the founding of Crisis Textline? Even though we all know, seem to know, like, "Oh, yeah, there's a new generation of kids who are more text-driven. It's really quite another thing to, I think, see that and see that in the data and see that in practice that they're actually able to reach out." And that what's the most interesting part about it is that it allows people accessibility to this kind of help when they normally wouldn't reach out. For somebody who's shy and apprehensive about calling, because that's how people are, in general, this day and age, that it's less threatening to text somebody. It's less threatening, so it'll encourage people to reach out and that they can do it without bringing attention to themselves. It's very hard to get to call someone without anybody else noticing unless you go into a corner, you go into a quiet space, but you can be in school or in class or at home or wherever you are if you feel that you're in crisis. Most people have cell phones nowadays, so it's something that they can immediately, that's there. Even though there's also tons or a lot of suicide hotlines and crisis lines to call, it feels less threatening to text, and so it's something that it absolutely spoke to me as like, "That is a cool idea." In addition to not only that's great that they're reaching out to teens in need that way, it's supplements, absolutely supplements. I don't think it competes, but supplements, the services are out there, but also has a great opportunity just as far as data is concerned. For all of these things that they're generating, all of this data that they're generating, all of these texts, they're able to actually come full circle, even though there is this year and a half old startup, non-profit, that they're already being data-driven, and that's something that was really compelling to me, just as a data scientist, that not only are they helping people, but they're already using data to help people is fantastic. Yeah, absolutely. Are you able to share any high-level statistics on the project? Yeah, so crisis text I launched in August 2013, and to date have had over 130,000 conversations and 6 million text messages, and so those have gone back and forth between the crisis counselors and the textors, and so I think by the time that this podcast airs, it should actually be around 6.1 million. For crisis text line, the point of storing and analyzing all this data is ultimately to make a change for those people who are texting in, and so it's not only just about the individual lives of the textor, but of the crisis counseling that they do, improving what's the mental health space is doing as a whole, and this has the potential to change so many different ways that we go about this today and handle this in our society, so this could affect education, it could affect public policy, and what other nonprofits are trying to do to further this kind of cause, and so by having this specifically as a digital data set in the way that they have it, it's far more agile. They can make a change even faster instead of waiting a few years for the researcher the data to become available. Yeah, I would imagine this is something too that can be looked at for any point of research rather than phone calls, which if they're not recorded, they're totally lost. So if you go to crisistrends.org, you'll be able to play with that some of that data yourself. So before we get into too much about the data, let me just mention crisis text lines privacy policy. So it's actually the same level of security and privacy requirements as the data sharing that's used by the CDC and the NIH. So they're always mindful even if you know what they're dealing with with this data that ultimately it is to help the services to help the textures. And so they want to keep the privacy and the integrity of what they're doing with the data at top of mind always. Yeah, it's great to see that level of standard. I don't know the CDCs myself, but I can't imagine that they're slouching on privacy. So copying that sounds like the right way to go. Yeah, exactly. So outside the data science community, this data can also inform folks like the public policy makers and researchers specifically about how teens experience crisis. And I don't think there's a lot of research that's been done at this scale or for this amount of time. And so what's great about this is that it's putting the data, this kind of data in the hands of experts and for the public who haven't had this in the past and could do something about it prevent stuff like this from happening in the first place. If you think about some of the studies in the similar vein like the CDC's Youth Risk Behavior Survey, for example, so that monitors a wide range of things of priority health risk behaviors. So imagine that survey which happens every and gets updated every couple of years could benefit from the type of data that CTL is seeing and the type of insights that they're gleaning. So we've mentioned earlier, you're approaching probably 6.1 million texts at the time. This will air over hundreds of thousands of conversations. That's a ton of data. How do you go about starting analysis and identifying patterns in that data set? Yeah, that's the first thing that you have to do with the data, right, is to start digging in and looking into the conversations and trying to see what patterns sort of emerge, right? So that's one of the things that we absolutely did put into our analysis was seeing if we can start picking out specific words that people are saying. One of the things that we had recently blocked about, for instance, was the use of the word "thank you." One of the things that they're always looking for is the quality of care that they're giving to the teens and to the textors. There is a post-conversation survey. So after every conversation they'll send a survey to the textor and say, "How do you feel? Do you feel better, say, more worse?" And so even though they have fantastic rates for that, that is not always indicative of whether or not people are actually feeling better if they don't say anything and coming with it. But one of the things we thought about is, "What about detecting gratitude?" Unless people are very, very polite, which I've heard is not very much not true of the younger generation. They won't say, if they come in in what crisis texts like in terms as a hot phase, they won't say, "Thank you," unless they've come down from that, if they're not grateful for the service or the conversation that they've had. So it was interesting to see that there's actually a very high rate. I think it was over 50% of all textors who come in will say, "Thank you," or will say, "Thanks," sometime in the conversation. And it's not in that post-conversation survey. It's not, "I feel better now, thanks," that gets sent. It's actually in the conversation to the crisis counselors, so they're directly saying, "Thank you." So it's interesting. Little behavioral things like that. Because we have texting data, we can start looking for the nuances in the language itself. And what's interesting to me about crisis text line data, as you said, they have this type of data. But for the typical hotlines, the phone lines, getting to that stage would be far harder. Again, from the data science point of view, man, trying to record voice data, first of all, recording voice data, convincing someone to store voice data, paying for that storage, for a nonprofit, then getting to a place where you have raw voice data, and even getting that transcribed as a first step. So you're basically back to where crisis text line is starting from. That's challenging, just to get to that stage, as well as getting all the nuances from voice data in terms of tone, in terms of pitch, is also really expensive in terms of time and effort. And so you can imagine that for quick wins, they've got so much sort of after fingertips in terms of the type of analysis that they're able to do. So when you came into the project, did you have a specific mission, or was it exploratory? A little bit of both. We had a three-month engagement. We wanted to focus in on one business problem, try to look around at their data and say, "What is it that we want to focus on?" So the focus of the study had started about conversation quality, like what makes a good conversation, and then it shifted into actually looking at all of our textures, not just what makes a good conversation, but really, who are we helping the least? And so this is where the analysis shifts to an understanding of when a texture becomes one who uses the service frequently without any change in their situation. So the repeat textures that you can see. So they have a statistic that 3% of their textures at one point were using 34% of the conversation minutes. So psychologists call folks who are in this type of situation, circulars. They're just circling, and they're the ones who require something like long-term therapy or counseling that crisis text line isn't equipped to give, and that's what the service is for. For these folks, it's not just about the immediate situation of going from a hot place to a cool place, but really, how do you help these folks break the cycle? And so without giving away the ending too much, so the crisis text line had a larger initiative of how do we help these folks who are in need and outside of what we specialize to do, but so my project was part of a larger initiative that they were doing, and once they've actually, they've already implemented a lot of changes, including the ones that came out of this project, and they were able to reduce that 34% of conversation minutes down to 8% of conversation minutes within a few months. And in these textures, ultimately, we're giving higher satisfaction ratings than they previously did. So the concern that maybe because they're going into other places, they're not happy, but it turns out that they actually were satisfied that they were given the alternative care. Yeah, it's really noteworthy the improvement, I would say. There are resources available out there for long-term care that they can say, like, this is all how do we actually help people, right? And so that's where this problem that we're trying to solve shifted. Well, how can we look at characteristics of repeat textures? How can we identify who they are so that we know how to help them in the future? And so that's where the analysis ended up going. But ultimately, the journey of getting there was also extremely helpful for crisis text line, like, because they are coming from a different space, like, for them, it was valuable to see how someone from sort of a corporate training, and somebody who, like, how analytics at scale would tackle. But as you mentioned, like all their data and this sort of problem, and seeing what that process, like, the exploratory process looks like. Like, how do you get from the raw data to features, like a feature space, then to the model, right? It's easy to say, like, okay, I'm going to make a model to determine whether or not people are repeating or not, or if I could determine how frequently people are repeating. But then the art and the exploration is getting from raw to that point, right? Okay, so there's two things that came out of the modeling efforts at the end of understanding of predicting repeat textures, right? One was, obviously, how soon you can predict someone is going to repeat. So we want to identify, you know, who becomes a repeat texture who is circling, and therefore needs other resources. Originally, the previous cutoff was actually around 20 or so, like, if somebody had come around 20 times, they'd be able to flag those textures so that they can start saying we really need to start recommending different resources for them, or we need to start changing the way that we're running our conversations. And we realized that it's actually far too high. What we got from the model was actually that they should make that cutoff far lower. So they've already implemented a change where they said it's not just 20, but we can start identifying these people at the fifth conversation, right? So immediately, you can imagine that that's already changing how they're handling textures and hopefully getting them the care that they need a little bit sooner. The second thing that we were able to do, as you mentioned, you know, we have a lot of interesting, really rich text data, is that there's text indicators to be able to determine whether or not people will be repeaters. One interesting one was the use of the word friend. So if you have a textor who says friend over a certain threshold, the language will say things like my best friends or I have a friend, my friends and I, that sort of a thing. And those are the, well, that's below the threshold is will be those types of messages. If you're above a certain threshold, suddenly the sentiment seems to change. It'll be I have no friends or my friends are all going or they're all leaving or they're avoiding me or they're abandoning me. And we realized that language is not that it sounds like people are just lonely, right? Are the repeat textures now coming to the service because they just feel like they've been abandoned by friends or they don't have any? Yeah. And then the other two interesting ones is that if you mention school, apparently that you're also become a repeat texture, and that when we started looking at the common engrams there, we wondered, I thought like maybe people are having issues with school and maybe that they are, there's so many school related issues, especially because we're looking at a lot of teenagers, right? Like what are all the common issues at school that people may be having? And it turns out when you look at that that the people are actually being in school is really an indication of who they are. They'll say I'm in middle school or I'm in high school or I'm at school right now or I'm going to school today. And so it's really more about location and who they are. And so it's really more indicative about the person's profile than it is about any particular problem that they're having. Although this is a service that's available for everyone, obviously this is their target demographic and maybe just how those people are going to be using the service is going to be different. And then the most serious one that we saw for a text indicator was the word hurt, right? So you see a lot of people who come in, I said you see a lot of different issues coming in. There's not crisis text. Unlike some other hotlines which are specific to suicide or abuse, this one is just anyone in a crisis. And so when you see words like hurt, you immediately think it could be a number of different things. Like are people getting bullied or people getting abused? Is this sexual abuse? Is this, you know, there's a lot of different things. And ultimately, when we see repeat textures and the word hurt, it's actually an issue regarding self harm or that they feel like they're being hurt. And so we see phrases like hurting myself or I want to hurt or it hurt. So with people who say that they hurt themselves, like people who are self harming, this could be, this could be very, it can be very much associated with clinical depression. And for those folks, they require more help than crisis text line can give them as a service, right? And so these folks can also be recommended quicker, you know, what are some services that they can take advantage of sort of locally. There's just the start, right? These, all of these little analyses are just the start of how crisis text line can look further and take these analyses and do more interesting things with them, as well as, you know, to already start changing how they are training their counselors to handle these particular textures. Yeah, so it sounds like that data set gives a lot of opportunities for a data scientist to come in and make a contribution. Tell me a little bit about some of the tools that were used. Yeah, so a lot of the tools that we ended up using are all open source tools. My background is in, I've spent a lot of time in parallel databases, so I used Postgres because it's something that one, it's, it's, there's a lot of commercial tools that are of parallel databases that are similar to Postgres that I've used in the past. And so use that as sort of a starting ground, like it has a lot of great extensions into, you know, things like PL Python, I think like Madlib, where you can use a lot of extensions to, let's just get it from a raw state to a process state. And then as well as using other Python libraries, I use a lot of like libraries that can then start picking up things like mood or modality, you know, whether or not things are a fact or an opinion. Like, so there's a lot of different things that you can pick up from text and sentiment that are available in those in sorts of libraries. Yeah, there's a lot of really great open source libraries and tools out there for doing natural language processing. One of the things I've noticed is that they tend to treat punctuation as a throwaway, like a place to split apart tokens. Yet in, I learned from your talk at Strata that actually you identified punctuation is incredibly important, not necessarily in the traditional sense, but in the use of emoticons and how those convey information, and that you actually don't want to strip those out in this specific use case. Tell me a little bit more about that process. One of the things I've learned about text analytics is that for every method and every canned solution, for every corpus that you deal with, you have to modify it somehow to be relevant to what you want to use it for, right? There's something in every corpus. And so for particularly for text data, it was almost like a no brainer. When you think about the words that people are using, one of the first things you think of is emoticons, right? So you have to make sure that there's two things in consideration. One, don't get rid of them, right? So if you just look at the types of characters that you want to strip out, like don't take away a colon or set my colon or parentheses. So I just had a list, like we had there's a bunch of ones, like lists of emoticons that are available. You can even download one from Wikipedia and just see what the most common ones are. And for the most part, people use like the typical like smiley and frowny faces. The weird ones with all the horns and stuff people don't actually use. But yeah, so one was making sure we retain all of those. It was interesting to parse those because not everybody will put a space between them, right? So it's a white space tokenization. You'll have to be a little clever about how you make sure that if people put a smiley face at the end of a word, you actually count that as two different words. And also that you don't just split on every punctuation mark because then you're going to split your smiley face into. The other interesting part of looking at punctuation is that we're also interested in whether or not people are asking questions. So for the counselors, do we want to know what ratio they have between their questions and their statements? And so those sorts of things, we also started asking and what realized like then you can't even just get you can't get rid of the end of the sentence punctuation either. So it was sort of twofold for us to figure out which types of punctuation we wanted to keep and make sure that our parsing rules were doing it in such a way that we were retaining all of those as well. So you had mentioned earlier that your work at Crisis Text Line was part of like a three month sabbatical. Is the work ongoing or are there ways other data scientists can contribute? With Crisis Text Line or with DataKIND? Well with both, frankly. With DataKIND, they do have ongoing data core projects. If this was a little bit little special because it wasn't it was just me on the project, but in general they do have every few months or so a group of nonprofit organizations who are looking for people who want to do stuff on a longer term basis. So you know kind of fit in the analysis on nights and weekends or when people can fit in the time. And you're also not alone. You also have a kind of a larger team, a group of data scientists who kind of see the project to fruition. So like absolutely, like if anybody's interested in doing this kind of project in general with a wide variety of nonprofit organizations, because they have a few that kind of crop up every time, they would have their choice. So they'd be able to go go check out DataKIND, go look at DataCore, board that. For Crisis Text Line specifically, I'd say that there's two ways that people can get involved. One of the main ways that they're actually looking to scale out their counselors because of the way that they've set up this great scalable platform, as I said, is giving a handle tens of thousands of conversations a day that they need people who can be remote counselors. So you don't have to be part of an organization or volunteer for a crisis hotline to be a counselor. They do have a remote training program. So yeah, like if you want to get involved, like that's immediately like the best way they can get involved. As I said, there's also, if people are just curious and want to look at the data themselves, they can do that as well. Excellent. CrisisTrends.org. I'll be sure to put a link to that in the show notes. So I feel a bit remiss that maybe I've mischaracterized this earlier when I kept saying, "Oh, it's for teenagers," because I imagine that Crisis Text Line is open to anyone at any age in a moment of crisis. So maybe we should take a moment and say how people could use this resource if it's something that they might need. Oh, you absolutely right. No, they don't turn, even though they are, as you said, they've obviously targeted towards teens. They don't turn anyone away and it's nationally available. So it's the short code 741-741. If anyone is, yeah, absolutely. If they're available 24 hours a day, seven days a week, you actually do get hooked up to a crisis counselor if you're in need. Oh, it's excellent. And how about at CrisisTrends.org, do they make sure they protect people's privacy? Absolutely. So if you go on to CrisisTrends.org, the data that they have is all aggregated. And so there's no PII, personally identifiable information that's available. All of that stuff is very, very protected and they want to ultimately want to make sure that they do right by the folks who texted and we are extremely mindful on how they handle that data. So while there's obviously some interesting things that are being mined, they are very protective and mindful of that. Oh, that's great. Well, I wanted to ask your advice on what you would say to a young aspiring data scientist, maybe a college grad, or even a high school student on how they can get started exploring this field. Oh, man. So the advice I give for folks who are interested in data science, first of all, that's fantastic. Go for it. And if you're in high school, really start making sure you take interest in your mathematics and statistics classes. And also, I'd also be mindful the way that you're using mathematics, statistics, and even programming in some of your science classes as well, right? So I know that it's not limited to just the math part of it. You don't have to be a mathematician because a lot of people who come into data science come in from various other disciplines, right? Like other disciplines are actually very, very technical. And I think a lot of folks don't realize how key that is to be to have a good solid math background. But and don't be scared of math. Math is not as scary as people think it is. For folks who are in their undergraduate degrees, and maybe even their graduate degrees, I'd also encourage them to make sure they have a good math background. Take as much take math. It's great. Make sure I know for a lot of mathematicians, they may not necessarily take a lot of programming, make sure that you know, study some programming, take a class in Python, and you know, see how it feels for you. And then there's a lot of stuff that you can do, like go volunteer at or and donate your time to data kind, like over a weekend and just see how much you like it, right? Like at your hands dirty. There's also obviously things like cattle, which will get people's hands dirty on with data and what it actually looks like. There's so many opportunities to, one, participate in something that's a little more organized, or two, just go look at data. There's so much available. And if you're really curious, you can also start generating your own data. I had a job on up before I broke it several times. But there's like that in your Fitbit that you can you can download all of that data and look at it, right? If you are curious about how you can make choices and how you can start, you know, playing with all of these things, all of that is it's not one, it's not scary. And two, it's easy to get started. Yeah, I think that's great advice. As far as data sets go, I actually have been wearing a basey span for almost two years now. So I've got a long time series of bio data on myself. And I've been thinking about open sourcing that maybe. Yeah, that'd be fun. That'd be exciting. I bet you get some great visualizations off of that data. Yeah, absolutely. So I like to wind up all my episodes by asking my guests for two recommendations. The first I call the benevolent recommendation, a link or nod to something that you think is worthwhile, but you're not directly affiliated with, and the self serving recommendation, something that ideally you get some direct benefit from by having it exposed here. Okay, so for benevolent recommendation, I really loved working with IPython notebook. Yeah, absolutely. So I think it's, and I'm sure you've had some people say this in the past, but, you know, obviously it's something that really helps tell the story, right? So be able to go to jump between code and results and graphs and all of those things all at once. I'd say that would be my benevolent recommendation to all the joy of all of my Pythonista colleagues out there, which would enjoy that. And then, so second, what is a selfish recommendation would actually be to, actually, and we've already talked about this entire thing to, you know, donate people to donate their time. But really, that would be my, that's my selfish recommendation for this, like, honestly, go. And as data scientists, like, go make sure you do donate your time, not just, like, if you volunteer, that's fantastic already. But go, go see if there's anything on data kinds, or any of the other data for good organizations. See if there's something that interests you, even if it's something that's, you know, you can go and donate like an hour or two of your time, but take the risk, get something that I think that if you're hesitant about, you should just go, go do it. And because it's really so, so rewarding in such a great cause. Yeah, absolutely. Well, one small correction, I think we have to start calling the IPython notebook Project Jupiter now. Oh, that's right. Thank you. Yeah, that's a fair point, fair point. But yeah, I use that on a daily basis myself. I think it's fantastic. I will anyway, and well, thank you so much for coming on the show. It's been great having you, and I really appreciate you taking the time to share your story with the listeners. Oh, it was a great pleasure. Thank you for having me. [BLANK_AUDIO]
