Pair Programming
Beyond the whiteboard interview
Edit
#interview
#minimum-viable-product-mvp
#ruby-on-rails
#software-architecture
#user-experience-ux
#pair-programming
Summarized using AI

Beyond the whiteboard interview

Joël Quenneville and Rachel Mathew • April 03, 2019 • Minneapolis, MN

The video titled "Beyond the whiteboard interview" features Joel Quenneville and Rachel Mathew discussing the interview process at thoughtbot during RailsConf 2019. The session aims to demonstrate a more humane and collaborative approach to technical interviews, moving away from traditional whiteboarding methodologies.

Key points discussed include:

- Preparation for the Interview: The nervousness of candidates is acknowledged, with an emphasis on creating a comfortable environment for discussing interview questions. A collaborative discussion model is suggested, as opposed to a confrontational one.

- Interview Structure: Quenneville explains that the interview will involve scenario-based questions that mix technical queries with consulting-related aspects such as team interactions and project scope.

- Technical Scenarios: The hypothetical project involves building a self-service appointment scheduling system for a gym. Candidates are prompted to explain their methods for understanding existing codebases and defining the Minimum Viable Product (MVP) based on client requirements.
- Onboarding Process: The candidate shares strategies for onboarding to a new codebase, including checking models and running test suites.

- Defining MVP: The discussion leads to defining the MVP, establishing the importance of self-service for gym clients while debating the complexities of scheduling appointments and handling holidays.

- Modeling Objects: The conversation evolves towards data modeling. Through iterative brainstorming, they explore various structures for schedules, appointments, and eventually arrive at a more sophisticated solution for handling holidays versus standard hours.
- Flexibility in Design: By the end of the discussion, they devise effective strategies that allow for scalability and the inclusion of features such as multiple gym locations and special events.

- Conclusions and Next Steps: The interview wraps up with a discussion about the next steps in the interview process, emphasizing the importance of collaboration, feedback, and continuous improvement in both interview and work settings.

Key Takeaways:

- The importance of a collaborative interview style that reflects real-world work scenarios.

- Understanding the nuances of MVP scoping and technical decisions within software development.

- The need for robust testing and modeling strategies that can adapt to the dynamic needs of projects.

Beyond the whiteboard interview
Joël Quenneville and Rachel Mathew • April 03, 2019 • Minneapolis, MN

RailsConf 2019 - Beyond the whiteboard interview by Joel Quenneville & Rachel Mathew

_______________________________________________________________________________________________

Cloud 66 - Pain Free Rails Deployments
Cloud 66 for Rails acts like your in-house DevOps team to build, deploy and maintain your Rails applications on any cloud or server.

Get $100 Cloud 66 Free Credits with the code: RailsConf-19
($100 Cloud 66 Free Credits, for the new user only, valid till 31st December 2019)

Link to the website: https://cloud66.com/rails?utm_source=-&utm_medium=-&utm_campaign=RailsConf19
Link to sign up: https://app.cloud66.com/users/sign_in?utm_source=-&utm_medium=-&utm_campaign=RailsConf19
_______________________________________________________________________________________________

You've spent a lot of time preparing for this moment. Your palms are sweaty. You take a deep breath, walk into the room, and shake hands with the candidate. Welcome to the interview!

Interviewing can be intimidating and our industry is notorious for interviews that are arbitrary, academic, and adversarial. How can we do better?

Come be a fly on the wall for a real interview! At thoughtbot, we've put a lot of thought into crafting an interview that is both humane and allows us to accurately capture a candidate's strengths and weaknesses relative to the real-life work they will be doing.

RailsConf 2019

00:00:21.279 Hi Joël, thanks so much for coming in.
00:00:26.320 Thanks for having me here for the interview.
00:00:32.320 Yeah, I'm really excited to get started. I know that you got an email talking a little bit about what our interview process is like, but I thought it would be good to go over it in person.
00:00:37.360 I'll give you a chance to ask any questions that you might have.
00:00:43.840 That sounds good. I'm feeling a bit nervous; I feel like I haven't busted out my whiteboarding skills in a little while, but I think I got this.
00:00:52.000 Yeah, interviews can be pretty stressful, and I know that whiteboarding interviews, especially, can be nerve-wracking because they're a little confrontational and the questions are often not what you would do every day as a developer.
00:01:00.160 But here at thoughtbot, we've put a lot of work into thinking about how we can make our interview more collaborative and closer to the work that you would actually be doing.
00:01:06.080 That sounds great.
00:01:12.159 Just to give you an idea of what we're going to do here, we're going to walk through a scenario, and I'll ask you a few questions. It's really an opportunity for us to discuss the questions and how you might approach them.
00:01:23.680 They're a mixture of technical questions as well as consulting questions.
00:01:29.600 Okay, I'm kind of curious. You mentioned consulting questions.
00:01:35.840 I work at a product company; I've never consulted. Can you give me a heads up on what those are going to be like?
00:01:41.200 Yeah, that makes a lot of sense. By consulting questions, I mean questions about product scope, team interactions, and processes.
00:01:46.240 These are things that are really important for a developer's day-to-day job but aren't necessarily about technical implementation.
00:01:52.399 Okay, so like Agile stuff?
00:02:04.960 That's definitely a big part of it, but there's also going to be stuff that goes beyond just process.
00:02:10.160 Alright, that sounds good. Let's do it.
00:02:20.800 For this first question, let's imagine that you're a developer at thoughtbot, and you're rolling onto a new project.
00:02:26.720 Your client is a gym in the area, and they have an existing code base that they're looking to expand upon.
00:02:32.319 They want to build a system that allows users to schedule sessions online.
00:02:38.400 Currently, all of their gym-goers are calling into the gym to schedule a class, which is not ideal.
00:02:44.640 So to get us started, how would you begin onboarding and getting to know an existing codebase?
00:02:51.680 I think I would probably look in the models directory first.
00:02:57.680 Assuming this is a Rails app, I would look for a user model since there's often a lot of logic in there.
00:03:04.400 I would also try to find the so-called 'god object' of the system.
00:03:10.080 There's often a lot of business logic centralized there, and I can see where some of the pain points are.
00:03:15.280 This would give me a good idea.
00:03:20.400 Nice, I think that makes a lot of sense. I like the strategy of finding the god object and then fanning out from there.
00:03:26.959 Are there other techniques that you might use?
00:03:32.319 Yeah, I think I might look at the routes file to see if things are sort of RESTful.
00:03:38.239 Actually, let me take that back. The first thing I would do is run the test suite.
00:03:44.319 I feel like that's the first thing I should say in a thoughtbot interview—testing is very important.
00:03:50.000 Yeah, we do like tests a little bit at thoughtbot. I hope you don't feel like I was fishing for a particular response.
00:03:56.000 I really just wanted to give you an opportunity to talk about all the different ways that you might onboard onto a project.
00:04:04.159 Now let's imagine that you've gotten to know the codebase and feel confident.
00:04:10.720 The client is itching to get started, and they are looking to thoughtbot for help defining what an MVP might look like.
00:04:16.000 I'll tell you a little bit about the requirements they've given us, and we can talk through the process of scoping an MVP.
00:04:22.240 Okay, so let's imagine that they're building a website. They want users to be able to book appointments through the website without having to call in.
00:04:30.800 They're planning on opening a new location sometime next year, so they imagine that this new location will have a different schedule.
00:04:37.760 They also run promotional events throughout the year. The next promotional event they're planning is sometime at the end of June.
00:04:45.280 During these events, they're going to have extended hours, but they sometimes do promotional events with shortened hours.
00:04:51.520 They're also closed on major holidays.
00:04:57.680 Given all of that, what do you think should be the MVP of this product?
00:05:05.120 I think when scoping an MVP, it's important to find what the core problem is that we're solving, and here it's self-service.
00:05:13.680 If a client can self-serve at the gym, then whatever minimum piece of software we build that allows for that is successful.
00:05:19.440 In that spirit, particularly given that the new gym will only be opening next year, we could probably cut that functionality and say we don't need support from multiple locations in version 1.
00:05:26.320 Okay, that makes sense, and I think that reasoning is solid.
00:05:30.160 Do you feel like being able to handle closures for holidays is core to the app?
00:05:38.080 That's tough. It adds a lot of complexity to the logic, and I'd rather not have to do it.
00:05:43.280 At the same time, it's hard to go to market without that.
00:05:50.400 If, say, you mentioned we'd be closed on the Fourth of July, if a client were to book an appointment for that day and then show up to the gym, it would be closed.
00:05:57.440 That costs us a lot in terms of goodwill and probably costs some actual dollars as well. It's just a bad experience all around.
00:06:04.000 Yeah, I agree. Having a client book a session, and then the gym is closed, is probably a really bad user experience.
00:06:10.080 Is there anything you could see changing about the requirements that would make you more comfortable cutting scheduling from scope?
00:06:17.440 Was that the wrong answer? Was I supposed to say yes?
00:06:22.400 No, I just think that talking about scheduling is interesting, and it might be worth discussing how this could change for you to feel more comfortable cutting it from the MVP.
00:06:28.080 Um, I wonder how often this project releases.
00:06:34.720 I ask because at my current company, we are on a two-week sprint cycle, and we release once at the end of every sprint.
00:06:41.279 So sometimes it can take maybe two sprints for something to go from idea to production.
00:06:48.320 Given that July 4th is pretty close, I feel less comfortable without that built into the initial release.
00:06:54.240 It makes sense that taking a month to ship something when you're trying to get out an MVP quickly would be stressful.
00:07:01.280 It doesn't mention it in the initial prompts, but at thoughtbot, we usually run one or two-week sprints.
00:07:07.520 It depends on the project and what makes sense for the team.
00:07:13.200 In this situation, maybe it would make more sense to have shorter sprints with a one-week sprint.
00:07:19.760 Okay, out of curiosity, have you ever worked on a project that used a continuous deployment approach?
00:07:26.240 Yes, and I think that's a really good point to bring up.
00:07:31.680 I do continuous deployment on my current project for really similar reasons.
00:07:38.080 The client is trying to ship a new set of features, and we deployed something initially that we wanted to build up quickly.
00:07:44.000 So I think that would work really well on this project.
00:07:50.400 If we had these tighter iterations, I would be more comfortable with a tighter MVP.
00:07:56.320 Those two things tend to go together in my mind.
00:08:03.760 I think that approach makes a lot of sense, and I'm glad you brought up these techniques around delivery that are also important for MVPs.
00:08:10.720 One alternative that we might consider—and this is something we did at my current company when we first launched—might be to go live with a beta version.
00:08:18.080 What we want to test and get feedback from our users is if self-service works and what things we need to learn from that.
00:08:25.920 Holidays are just a special case that will take time to gather feedback.
00:08:33.440 If we were to ship a version that doesn't have holidays in it, but maybe only to selected users and communicated with them that this is beta software,
00:08:42.000 and that it allows booking any time—even on holidays—then we could send out an email reminder a week before, saying the app will allow booking on July 4th.
00:08:50.560 Just a reminder that we are closed that day.
00:08:57.840 Yeah, that would allow us to gather user feedback and get the iteration loops tighter, while not having the pressure of shipping to people on July 4th.
00:09:04.960 I think that's a great idea. I also like that because when you're putting out an MVP, there might be some user experience issues or small bugs.
00:09:11.680 Having this beta version sets expectations for users that this is going to get better and it's normal to have a few rough patches.
00:09:19.760 I feel like, with that, we really talked over what an MVP would look like, and we can maybe start thinking about how we would implement that.
00:09:25.440 That sounds good. Just to reiterate where we landed: we want to initially build out a flow that allows a user to go online, book an appointment, and we're not going to worry about any flexible scheduling.
00:09:32.640 You can say the gym just has nine-to-five hours, and we also won't worry about multiple locations.
00:09:40.000 So given all of that, how would you go about modeling these different objects and how they're related to each other?
00:09:46.079 I think it would have some kind of Trainer object and then a Schedules table.
00:09:57.760 There would probably sit between the two, maybe like a Slots model or something that could belong to both a Trainer and a Schedule.
00:10:05.839 I guess we'd have to probably pre-generate those ahead of time.
00:10:11.760 Actually, can I take that back? I can see where this is going, and I actually did this on another project, and it didn't turn out too well.
00:10:20.000 This was a payroll application I was building, and we leaned too heavily into Active Record associations for all our modeling.
00:10:28.640 Core to that was the concept of a pay period and everything associated to a pay period.
00:10:35.680 This meant I had to pre-generate them every month, and then we hit a daylight savings boundary and things got weird.
00:10:41.840 We ended up with hours associated with the wrong pay period, or maybe orphaned, or not associated to a pay period at all.
00:10:48.000 We ended up with a lot of issues with our payroll reports, which is not something you want on a payroll system.
00:10:54.160 This bug wouldn't die; every few months we'd run into another variation of it and think, 'Oh, the pay periods strike again.' So, no, I'm going to stay away from pre-generating those things.
00:11:02.560 Okay, I'm sort of trying to visualize this in my head.
00:11:09.760 When I think of a calendar, I probably don't want to model the white space on the calendar but rather the individual events on it.
00:11:16.000 So, what if instead of having that table that sits between the Trainers and the Schedule, we have a table that represents appointments?
00:11:22.080 These are slots that have already been taken, and when we want to show what's available for others to take, we can calculate that at runtime based on what we know has already been booked and the current schedule.
00:11:30.560 So just to go over, to make sure I understand your solution, we would have a Schedule table with the hours associated with a number of appointments.
00:11:36.960 Trainers would also have many appointments, and an appointment would be for a particular Trainer at a given time.
00:11:43.680 That really nicely covers what our MVP would look like.
00:11:50.720 So, if we start to layer in some of the features we talked about initially, let's imagine we've built the MVP, the client loves it, and comes back to us.
00:11:57.040 Now would be a great time to make the system handle special events.
00:12:03.680 This could include extended hours for a promotion or being closed on a holiday like July 4th.
00:12:10.080 How would you add that modeling into our current solution?
00:12:17.440 I'm glad we decided to put this off earlier because this is kind of the gnarly part.
00:12:24.240 Previously, we had the Schedule table that just had a single row in it, which feels a little weird.
00:12:30.080 Now I think we could add a start and end date columns to it.
00:12:35.920 This way, we would have multiple rows in this table, and a particular schedule would only be active during a specific time period.
00:12:44.320 One nice aspect of this is that it allows for setting up schedules ahead of time. For example, we could have a winter schedule and a summer schedule.
00:12:50.480 The admin could go in and set these at the beginning of the year, saying what the hours are for winter, and what they will be for summer without having to make changes now.
00:12:56.160 That nicely covers a common use case for gyms that wasn't talked about in the initial requirements—the scheduling over large blocks of time.
00:13:05.760 Could we talk through how an admin would go in and close the gym for July 4th?
00:13:12.160 I think of a holiday as being like a micro version of a winter and summer schedule.
00:13:20.320 I guess an admin would go to the admin panel. We already have a summer schedule set that goes from June 21st to September 21st.
00:13:27.200 They would then have to end that schedule early on July 3rd, create a new schedule that's only for July 4th where the gym is closed,
00:13:34.720 and then start another schedule from July 5th through September 21st. That schedule would need to be the same as the summer schedule.
00:13:43.120 I feel like that might be a little error-prone for admins that would have to recreate this old schedule.
00:13:50.080 It is somewhat clunky. Maybe I could add a cloning button to copy the other half.
00:13:56.320 The cloning button would save time, but still, it's a lot of work to ensure our models support the scheduling.
00:14:03.040 I liked what you said before, where you mentioned how we handle schedules for different parts of the year.
00:14:09.760 It seems there's a logical separation we're not accounting for.
00:14:15.760 You mean that a schedule and a holiday are similar but not exactly the same?
00:14:21.760 Yes, they feel different to me.
00:14:27.040 Now, going back to visualization mode, I previously saw a calendar where one big event covers the whole summer.
00:14:33.040 Then, when adding an event in the middle, I would just break it into two and put a new one in.
00:14:39.200 But I need to think of it as having one big event that covers the summer and overlaying a holiday on top.
00:14:46.080 What if I model it this way? I could have a table that represents holidays as a separate entity, while keeping the schedules as previously described.
00:14:53.920 Now, when looking up availability on July 4th, we check the holidays table for any applicable overrides.
00:15:02.240 If there are holidays today, then the holiday schedule applies. If we're on July 3rd, there are no holidays, and we fall back to the current active schedule.
00:15:09.760 That feels like a more elegant solution, especially in terms of UI.
00:15:16.000 With this method, holidays are distinguished from schedules that administrators enter into the system.
00:15:22.240 I think we only have one feature left from our initial prompt, which is to support multiple locations.
00:15:30.560 If you were going to modify our existing database table diagram?
00:15:36.720 I guess the obvious addition is a Locations table now that we have multiple gyms.
00:15:43.680 I probably want an association to schedules since schedules are now per location. They don't all share the same schedule.
00:15:50.720 I think I would want to implement validations on the schedules table to ensure there is only one active schedule at a time.
00:15:57.680 With multiple locations, we'll need to scope that validation.
00:16:04.440 Also, I would need to add testing around that because it could introduce bugs.
00:16:11.360 There could also be an association to holidays, but I’m not sure if we have different holidays at different gyms.
00:16:18.640 That seems a little bit weird. However, the promotions might be different.
00:16:25.840 I could see different promotions targeting different groups of people at different gyms.
00:16:32.240 It seems like our solution expanded nicely out of our flexible scheduling system.
00:16:39.240 I think that's all we need to add—just the Locations table.
00:16:46.720 That confirms we made the right choice in the previous question—our architecture is now very expandable.
00:16:52.640 Great! I’ve gone through all the questions I had for you. Do you have any questions for me?
00:16:59.760 Yeah, I'm curious about the next steps for me after this.
00:17:05.760 That's a good question. Usually, there are two more steps after this.
00:17:11.520 The first being a code review. You'll receive a link to a GitHub project with an open PR.
00:17:17.680 The project is based on one of thoughtbot's internal projects, and you'll be asked to leave feedback and questions.
00:17:24.440 Try to review it the way you would approach any normal pull request.
00:17:32.000 After that, we have a pairing interview. You'll come in—we try to do them on Fridays.
00:17:39.680 Thoughtbot has internal time where we work on investment projects.
00:17:46.000 You'll pair with someone in the morning, have lunch with the team, and then pair with someone else in the afternoon.
00:17:53.440 Does that mean I might end up pairing on Factory Bot?
00:18:01.200 Yeah, a lot of our Friday projects, which are often open source, are born out of ideas we started on Fridays.
00:18:08.240 It's a very real possibility. It’s also very cool.
00:18:15.840 I'm curious; you mentioned earlier there’s a pull request review phase.
00:18:22.560 I've not seen that in interviews before. Is there anything I should watch out for?
00:18:29.080 No, similar to this interview where we want it to be discussion-based, we’re just looking for how you might approach these problems.
00:18:35.760 During the code review, we also want to see what you look for and how you communicate your feedback.
00:18:41.680 Code reviews are an important part of development, and we’re curious how you approach them.
00:18:48.640 I like that you emphasize team collaboration and how that works.
00:18:54.080 Yeah, it’s really important at thoughtbot too. Do you have any other questions?
00:19:01.520 Could you walk me through what a week in the life of an average thoughtbot developer is like?
00:19:06.800 Sure! From Monday to Thursday, I work on client work.
00:19:12.560 What that means for me right now is that I'll pair in the morning with another thoughtbot developer.
00:19:18.440 After lunch, I'll continue whatever we're pairing on by myself.
00:19:24.880 Around three o'clock, we take a coffee walk, which usually for me doesn't involve coffee.
00:19:32.400 It's just an excuse to go outside and get some fresh air. I find that when I take time away from the computer, everything sorts itself out.
00:19:38.960 When I return, all my problems seem fixed.
00:19:45.680 On Fridays, I've been working on my RailsConf talk.
00:19:52.000 Congratulations! What’s the topic?
00:19:59.120 We're discussing our interview process, where we've put a lot of work into making it more collaborative, closer to the work that you might actually be doing.
00:20:06.880 We're hoping that it's useful to others.
00:20:13.680 I have to say it’s been one of the more enjoyable interviews I've done, so I'm glad you're going to share that.
00:20:19.920 Thank you; it's great to hear that feedback. Good luck!
00:20:27.680 I think with that, we're all set.
00:20:34.560 Thanks for having me here.
00:20:39.040 Thanks so much for coming in.
00:20:46.160 Now that Joël's gone, it's probably a good time to fill out our interview rubric while it's still fresh in my mind.
00:20:54.080 I'm a big fan of this rubric; it helps me put different parts of the interview into perspective.
00:20:59.680 Okay, all these questions are asking me to identify what level I think Joël might be at—expert, practitioner, apprentice.
00:21:07.520 For the first question, was the candidate able to talk through different strategies for integrating onto a new project?
00:21:12.320 Joël had a lot of great techniques for getting to know a codebase.
00:21:18.720 He didn't discuss enhancements he might want to make, like fixing bugs in onboarding docs or fixing test warnings.
00:21:24.880 However, I didn't prompt him for that after he finished talking about getting to know the codebase.
00:21:31.360 So for this question, I think he's probably a practitioner.
00:21:39.760 Was the candidate able to define a project product and justify their decisions?
00:21:46.080 I really liked how Joël answered this question.
00:21:52.960 He talked through the features he would keep and why, along with how deployment affects the MVP.
00:21:59.760 That’s very important, and all candidates mention this, so I think he’s definitely an expert here.
00:22:06.720 Was Joël able to model complex objects in a domain?
00:22:14.800 He quickly proposed a good solution for our MVP.
00:22:21.040 Joël had two approaches and used his real-world experience to define which way he would go.
00:22:27.760 He also showed a solid approach to the scheduling component.
00:22:34.720 While his initial solution was a bit clunky, when we talked it over, he refined it to a better fit. I'm going to classify him as a practitioner.
00:22:42.080 For the last question, was the candidate able to talk through the pros and cons of their technical solution?
00:22:48.880 Joël did well here too.
00:22:56.080 When asked how an admin would enter a holiday, he could quickly identify the disadvantages of his solution.
00:23:02.640 Even if he had a bit of difficulty coming up with a more elegant solution, I think Joël is also an expert in this aspect.
00:23:09.200 Overall, Joël did really well in this interview, and I think we should consider hiring him or at least bringing him in for a code review.
00:23:14.960 I’m going to give him a call.
00:23:22.080 Bringing these elements together, I’m excited!
00:23:27.680 Hi Joël!
00:23:35.680 After our interview, I felt really good about how you did.
00:23:41.440 I would love to have you come in for a code review.
00:23:47.520 Okay, perfect!
00:23:53.040 So I think with that, we are all set.
00:23:59.200 I hope this has been helpful in seeing how we do things here at thoughtbot.
00:24:04.400 If you think this is a better way to interview, you should try going through it yourself.
00:24:10.320 You can apply by going to thoughtbot.com/jobs.
00:24:16.000 Feel free to reach out to me or Joël.
00:24:23.920 I don't have a prominent social media presence, but Joël’s Twitter is on the screen.
00:24:30.880 Thank you so much, everyone!
Joël Quenneville
@JoelQ
Rachel Mathew
@ronarachel
Explore all talks recorded at RailsConf 2019
+102