Meet the Makers #1: Daniel Pyrathon (Ranger Dave Bot)

Meet the Makers is a weekly series of interviews of bot-makers by bot-makers for bot-makers. This week, Arun Thampi, Founder at getbotmetrics.com interviewed Daniel Pyrathon who built a Bot for the Outside Lands music festival called Ranger Dave.

If you know a bot maker that we should interview please send us a note at: botmakers@getbotmetrics.com

Arun Thampi: Hi, everyone. My name is Arun. I’m the co-founder of BotMetrics, an analytics and engagement platform for chat bots. Today we are speaking with Daniel, who built a very interesting bot called Ranger Dave for the San Francisco Outside Lands Music Festival we’ll be talking with Daniel to kind of find out how he got into bots, what Ranger Dave does exactly, what the future plans for it are, and why he’s excited about the chat bot space in general. Hi, Daniel.

Daniel Pyrathon: Hi. Thank you very much for having me.

Arun Thampi: Cool. Before we get into the bot itself, maybe you can spend a little time talking about who you are and probably a brief introduction about yourself?

Daniel Pyrathon: Sure. Thank you very much. My name is Daniel Pyrathon. I currently work as a back end engineer for a company called Coffee Meets Bagel, based in San Francisco. As far as my background, I graduated with a degree in computer science from the University of Plymouth in the U.K.. I’m very, very passionate about open-source. Probably one of the most notable things, in my opinion is my contribution to Jango as part of Google Summer of Code 2014, where I re-factored the meta APIs. Then, on top of that, [inaudible 00:01:27] I moved to San Francisco just like, a year ago, so I’m fairly new, but I just love being involved in everything. Two probably other passions that I have now is automation and productivity, and that’s where the big bot picture fits in, in most of my work, let’s say.

Arun Thampi: Got it. You’ve kind of been building bots, I’m assuming, on Slack as well as on the consumer side?

Daniel Pyrathon: Yes.

Arun Thampi: That’s cool. Cool. Maybe you can talk a little bit about what Ranger Dave does, who is it for, and what platforms it’s on?

Daniel Pyrathon: Sure, sure. Ranger Dave bot is a virtual assistant for San Francisco’s Outside Lands Music Festival. Just to put it into context, Outside Lands is a 3 day festival in the heart of San Francisco, in the Golden Gate Park. From my understanding, one of the biggest music festivals in California, if not more than that, and it features really, really good, really good stars. This year we had J. Cole, LCD Soundsystem, Lionel Richie, really exciting people.

What my bot does, it provides a sort of virtual assistance, a presence which knows everything and everyone in the Outside Lands, which will be able to answer questions about [inaudible 00:02:53] FAQ, lineup, and for example, artists. It can guide you through your musical tastes, your food choices, and if you ask the right questions, it also leads you into the Outside Lands Speakeasy. That’s sort of like the [crosstalk 00:03:09]

Arun Thampi: Interesting. Just a little Easter egg in there.

Daniel Pyrathon: Yeah. As far as let’s say, who is it for, I’d probably say that it’s got 2 main sort of uses. First of all, probably the most obviously is the Outside Lands attendance. The Outside Lands attendance, as part of the festival joy, and most of the times, they have simple questions that they want answered, like for example, “Where is the restroom?” or “How do I get to the stage?” “When is Lionel Richie playing?” The problem is that most of the time what they do is they look for their booklet, which they probably lost somewhere. Then they turn around, they ask their friends, and then they don’t know what to do, so they pull out their phone, and they start searching.

That is kind of like the most inefficient thing ever, because from that point of view, most of the times, you’re searching for something. You’re in a festival, and the last thing you want to do is start looking, searching for information. Probably there’s going to be a lot of congestion inside there. Most of the time, it’s going to be highly inefficient. What my bot does, what our bot does, because it’s part of the Hack-a-Thon, and I’m part of a team. What our bot does, instead of letting the user search for information, the user can just ask for information and get a response. It has that really human aspect, which is fundamental, first thing.

Second thing, it’s also going to have far less bandwidth requirements, because we’re actually only dealing with text. You don’t have to load numerous ads and pictures and stuff like that, which you’d probably do in a web search. This is probably the biggest use case.

Then we have, also, another use case, which is very, very important in my opinion, which is that the festival organizers were also engaged with the users. [inaudible 00:05:11] engaged with attendants. They want to give information, and entertain, and in any case, have a line of communication with their mass, with the people. I think Ranger Dave there is really important, because it allows that sort of 2-way communication between also the organizers, to the mass.

Arun Thampi: Right. Was there both a NLP component to the bot, and what powered that NLP?

Daniel Pyrathon: Sure, sure. Ranger Dave boy couldn’t exist without 2 main components. The first one is obviously the messaging platform upon which is Facebook messaging API. The bot is hosted on Facebook, and the reason we chose Facebook is because first of all, it’s intuitive, right? It’s something that we all know how to use. It allows us to give … To give a structure in messaging, so you can provide multimedia content like the cards and stuff like that. Finally, it’s also universal. Everyone knows how to use Facebook Messenger. Most of the people have it installed. You don’t have to install something new. It’s one click, connect and you just chat away. Once you’ve finished, if you want to dispose of the bot, you just delete the conversation. This is the first component.

Second component, which is fundamental, which is the human aspect, is the NLP component. We chose Api.ai … Actually, my teammates suggested Api.ai and that has probably been one of the most winning components of the bot, because it provides a very, very simple and intuitive API infrastructure for the bot to integrate. Most of the dynamic conversations wouldn’t be possible without using Api.ai.

On top of that, we also implemented a small wrapper of our own NLP around some of the things. For example, the Speakeasy component, you actually have to keep track of which riddles the user has resolved. He cannot resolve riddle 3 if he hasn’t resolved riddle 2, for example. At that point, we actually had to implement our own logic around the Api.ai, which was fairly easy. One thing which I really found beneficial is to keep things as modular as possible, because that will allow us in the future to, for example, take Facebook out and put SMS in, for example.

Arun Thampi: Got it, yep, yep. Any interesting secrets or tips that you learned while building the bot that you’d like to share with other bot makers out there?

Daniel Pyrathon: Sure. I really would love to share my experience. First of all, I have very, very little experience in this area, but just by putting this [into 00:08:17] production, I realized so many important things. The first thing which I would suggest anyone who’s actually [inaudible 00:08:24] writing a bot is, log everything. Most of the bugs that we found were not implementation bugs, but they were actually bugs based on our assumptions, based on our assumption of how people would interact with the system. By logging everything, we were able to reconstruct conversations, we were able to understand what the users actually wanted to do, and what the system did by mistake, for example.

That’s been by far the most priceless thing that we’ve learned. That has also allowed us to integrate a lot faster on the product. Once we actually started logging, we understood things much easier and we [inaudible 00:09:03] on top of that. I suggest, for example, we deployed on Heroku, so paper trail is a very, very simple option. It’s a one-click install and it allows you to aggregate content. It’s quite cheap. That’s one thing.

Second thing is, terms of service. At the end of the day, our bot relies on our messaging component to work, right? We obviously have to really honor the terms of service that Facebook gives us. I really encourage any person who puts a bot in production that uses, for example, the Facebook API to read the terms of service, because there are some things that you cannot do, you have to make sure you don’t do them, because if you do do them, then your product is basically garbage. Facebook is just going to flag your product.

Arun Thampi: What are examples of things you can’t do?

Daniel Pyrathon: Sure. Let me give you an example. Ranger Dave bot was a product that came out of a Hack-a-Thon, by which [inaudible 00:10:01] Outside Hacks, which is the official Outside Lands Hack-a-Thon. We were one of the winners, but there was also another winner, which was an application called United Light, that we wanted to, because we had friendship with them, we wanted to also publicize their application, right? Throughout the festival, the [inaudible 00:10:25] thing we did is, because we had reference to all the users connected to the bot, we decided to broadcast a message saying, “If you’re excited about Outside Lands, and Outside Hacks, look at this other Outside Hacks winner, because it’s a really cool project.”

We realized 10 minutes after that we had violated one of the most important terms of service of the Facebook API, which is basically, you should not do mass publicity. Luckily, we did it once and we didn’t get caught, put it this way, but we learned from it. Luckily, not the hard way, but I really encourage people to understand exactly what is possible and what is not. Your app depends on it. Unfortunately, it’s a single point of failure, I’d say. If Facebook cuts us off, we’re basically out of the game.

Arun Thampi: Great, right. That’s very good tips. The first point that you mentioned about conversations being completely random is something that, coming from a web and mobile world, it takes a little time to get used to. Logging everything is definitely great advice.

Daniel Pyrathon: Yeah.

Arun Thampi: Sorry, go ahead.

Daniel Pyrathon: Also, one thing I really learned from experience is actually context relies … Obviously, it relies a lot on time. What I mean by this is that context is really fundamental. If we were speaking about something, you have a specific intent, and maybe what I was speaking about 2 hours ago with Ranger Dave is not the same thing that I’m speaking about now, so ensure the conversations are [or 00:12:12] finished, or or completed, or, after a sufficient amount of time to process, for example, 5 minutes, mark the conversation as complete.

One of the things that we found with logging is people would write to Ranger Dave, asking a completely different question, but our system would think that it’s related to the previous answer, which wasn’t solved.

Arun Thampi: Got it. Yep, that’s great advice. Tell us about the launch itself. What did you guys do to promote it? Obviously, I’m sure Outside Lands themselves helped with a lot of promotion, but what kind of tips can you share about promoting your bot?

Daniel Pyrathon: Yeah. We’re actually super lucky, because even though we won the Hack-a-Thon, we really worked a lot to make the product even better, better, better. Obviously, Outside Lands featured us as part of their digital experience. Outside Lands has a App Store iPhone app you could use, and inside that app, they actually put a link to our project. We’re super blessed from that, and obviously that grew the number of unique users drastically. The other things which we found really engaging is also showing, through social media, Twitter, and some of the Hack-a-Thon [peeps 00:13:44].

We always show that we were improving. I think that’s a really important thing when you’re building a product, right? We launched our first version, which is basically just like a hack. Then, we showed that we’re actually [inaudible 00:13:56] on top of it, making it serious. We basically re-structured the entire infrastructure, and we made it scalable, because it wasn’t at all before. I think people noticed this. Obviously, most of the growth in terms of consumer growth, came from the Outside Lands itself, Outside Hacks. Once we were there in the Hack-a-Thon, in the festival, we were also able to show our product around, show our enthusiasm.

Just to put it into context, we had around 800 unique guests which interacted throughout the entire festival, and more than 3000 unique conversations were actually concluded. What I mean by that is, the intent was actually 100% completed. Our [inaudible 00:14:45] was basically successful.

Final statistic, which I’m kind of proud of, is 50 people … Actually, I thought it was much less, but now we realize that it was more, around the lines of 40 or something, 40, 50 people also passed the challenge, the Speakeasy test and got access to the Speakeasy, which was free beer, so, you know. Highly beneficial.

Arun Thampi: Yeah. That’s great marketing that you kind of … That was part of the bot. That’s cool.

Daniel Pyrathon: Yeah, yeah.

Arun Thampi: Any key lessons that you learned once the bot was being used by real users? What were some of the lessons learned? What were some of the challenges that you faced, too? One of, I guess, the challenges as well as the beauty of a bot is that it’s completely open-ended, like you mentioned before. People can ask it anything. Were there some interesting lessons or insights that you learned?

Daniel Pyrathon: Yeah, sure. Sorry. One of the biggest lessons that I learned is that traffic congestion is a lot worse than I expected. Inside a festival, I knew it was kind of a problem, but I didn’t realize it would be such a big problem. Unfortunately, we did realize that traffic congestion did affect the user experience on some aspect. Not in a drastic way, but it does affect the usage of the bot.

One thing I’d have to suggest there is if you’re building bots for large crowd, ensure that your intents can be finished as fast as possible, because people are going to be waiting for … it’s going to have a certain level of lag. Luckily, things are getting better, but that was one thing. Probably a suggestion that I can give there is, try to make your intents as small as possible, let’s say, computable in as small time as possible. Also, most importantly … I guess this is kind of common software engineering principle, is design as modular as possible, because at some point you’ll be able to take that messenger API away, and plug in an SMS, or maybe embed it inside the native app of another festival, let’s say.

By designing modular, you’re able to kind of overcome things by providing more solutions for people. Maybe you’ll have an SMS number that you can text to speak to the boy. If Facebook is working for you, then you can use Facebook, or you can find other mediums. By increasing the number of offering you have horizontally, you’ll have more probability of engaging with more users, right?

Arun Thampi: Right.

Daniel Pyrathon: That’s, I’d say, one of the main key points. The second thing, probably secondly, [inaudible 00:17:47] the first, is, as much as I love people interacting with a bot, a successful bot is a bot that gives you an answer, and allows you to do other things, allows you to carry on with your life. In this case, I want my bot to answer the user’s question, so the user can enjoy the festival more, right? I guess, just as I want it, also the festival organizers want this. Make sure that your intents are as simple as clear and concise as possible, and answer the user with the best, intuitive way possible.

Arun Thampi: Yep, yep, that’s solid advice, because that’s kind of where bots win over mobile apps, is because the whole interaction from having to install an app is more difficult than starting a conversation with a bot. It’s all about how you can enable the user to get to their end goal as quickly as possible.

Daniel Pyrathon: Yeah, yeah. Yeah, I agree.

Arun Thampi: That’s cool. What are the next steps for you with the … Is there a plan to kind of take this bot to other music festivals, or other events, and so on?

Daniel Pyrathon: Yeah. We want to take the learning from this festival and [inaudible 00:19:03] in top of the product and improve it. We’re really happy with the way things went. Really, really satisfied, but we want to be able to improve on the product. I personally believe that I can … Us as a team, we’re very sort of horizontally focused. We’ve all got really good skills that complement each other, so we can definitely, with more time, improve the product and make it much better.

The plan, I’d say, for me, for us, is to re-brand the bot and propose a much better version for other festivals. We still haven’t narrowed down, but definitely in the next days, we’re going to start understanding who would be interested, based on the success story we had at Outside Lands. We’ll try to re-propose the same sort of offering for other festivals and see how that goes. The main take home for me is that attendants do benefit from a virtual assistant. We’ve seen this from the number of users that have interacted, but also by reading the conversations. We understand that they do really benefit from it, and they do gain value from it.

The second point is, festivals, on the other side, festival organizers can communicate and engage with their attendants with a bot. That’s also a clear, winning step, if they want to provide some kind of [game-ifying 00:20:40] aspect, or some kind of entertaining game, riddle, for the attendants. Another thing, for example, is lost and found. People will be able to interact with a bot if they lost something, and the bot would probably kind of interact with the organizers.

Finally, one thing which, it’s kind of on the top of my head is, we’re building separate experiences every time for the users. We want them to be unique, but another thing which I’m thinking of is, what about building a kind of framework, something a bit more generic, like sort of a bot as a service? That’s kind of like where my head is going now. I need to sort of mature the thought, see if it’s actually a complete idea. This could be sort of a direction to it at some point.

Arun Thampi: Yeah, definitely. Like I mentioned earlier, I think it makes a lot of sense for people to interact with bots at these events, because I never download apps, for whenever I go to events such as these, but just this kind of thing that you can just speak to on a messaging platform makes way more sense than an app. I’m definitely excited to see this kind of be a trend among more events and festivals and so on.

Daniel Pyrathon: One thing which I really found fundamental, like the fundamental difference is that, if you’re using an app or if you’re searching on the web, you’re always searching for stuff, right? You’re actually actively looking for an answer. With a bot, you just ask, and you should get the answer, if the bot works, obviously.[inaudible 00:22:29] if the bot works.

That is a completely different experience. It’s a completely different experience. You ask, and you get the answer. You don’t have to start there, searching for stuff. You don’t want to do that. You want to get the answer now. You wanted to get the answer yesterday, basically. You want to get the answer as fast as possible. This, connected with the multimedia features that, for example, Facebook offers, with [these 00:22:55] personalized cards, give a whole new level of experience, which is, for me, a clear winning point.

Arun Thampi: Yep, yep. For sure. Cool. Thanks, Daniel, for sharing all of the great lessons that you’ve learned. Where can we find this bot, the Ranger Dave bot?

Daniel Pyrathon: Sure. The Ranger Dave bot can be found on Facebook. I can provide a link, but it’s Facebook.com/RangerDaveBot. I’ll send you a link after.

Arun Thampi: Cool.

Daniel Pyrathon: All you’ve got to do is pick the “send message” and you’re going to automatically be connected to the bot.

Arun Thampi: Cool. Where can people find you, on Twitter, and get help and so on, because you’re doing interesting work, both with open source, as well as projects such as this.

Daniel Pyrathon: Thank you. Thank you. I’m kind of find-able anywhere, really. Daniel Pyrathon is my name. My handle for everything is PirosB3. I guess you can find me anywhere that way.

Arun Thampi: Cool. Awesome. Thanks, Daniel, for your time. We’re looking forward to …

Daniel Pyrathon: [crosstalk 00:24:10] if you think there’s something that can be improved on the product, please let me know, because I’m really happy to improve it and bring it a step ahead.

Arun Thampi: Awesome, yeah, for sure. We’ll be doing that. I hope the readers and the audience will do that too.

Daniel Pyrathon: Yeah, yeah.

Arun Thampi: Cool. Thanks, Daniel, for your time.

Daniel Pyrathon: Thank you for having me.