Ready, set, data: Perfecting the art of talent matching with AI

Mark: Hi everyone. Thank you for coming. My name is Mark Oretta. Brief history about me - I spent a decade at a media company called iHeart Media. They have a kind of small festival around here if you've ever been. Then I did a couple of startups, one went public and one did not go public. And then now I've spent the last four years at a company called JobTarget.

JobTarget is a company that you probably have never heard of but have very likely interacted with. We are a recruiting advertising platform for employers to help them manage their job, their job advertising, their job descriptions, find job seekers and then shepherd the job seekers through the hiring process.

Our most popular product is our programmatic advertising product which uses millions of signals to manage these advertising campaigns in order to optimize the ad spend to make sure that our employers are getting the best candidates they can for the most economical price that they can.

So we are an integrator. We connect sites like Indeed, LinkedIn and ZipRecruiter to applicant tracking systems. I'm sure if you've ever looked for a job, let's say you wanted to find a job at a company like AWS, they would post their job on aws.com/softwaredeveloper and they would post their job and then they would use a company like JobTarget to then distribute that job across the internet.

And then if you decided that that job was the job for you, you would apply and when you applied, it would come into our system, we would process your job application, we would enrich it if that's what the customer wanted and then we would send it over to the employer for them to hopefully hire you since you were so excited about it also.

Vic: Thank you, Mark. My name is Vikram. I head the delivery for Americas and AWS globally for anything related to data, analytics and machine learning services. We are a global company and anything related to data comes from our organization.

When it comes to Rackspace, we are one of the premier partners in AWS ecosystem. Our technical accreditations showcase how credible we are in the AWS space - we have over 2,800 AWS acres, over 200 plus professional certifications. And then importantly, we have over 16 AWS competencies and getting a competency is not an easy thing, you need to deliver, prove yourself and then more importantly, do it in a rinse and repeat it multiple times.

On another hand, we have over 16 service delivery designations as well and we're planning to add 4 to 6 more next year and we continue to grow from there on as well from the industry research perspective named as leader in four different things. And one of that is AI and machine learning services as well.

Mark: When I joined JobTarget about four years ago, we were completely on prem and what we were realizing was that our business was becoming very, very limited by our previous technical decisions that were made long before me.

We were running into scaling issues. It took us weeks to get a deploy out. I'm sure, you know, if you've been around for a while, you knew what that used to be like, you would have an outage window on the fourth Monday of every month for three hours and you put up the little pretty sign and then you would do the thing.

And then my devs were also always chasing bugs, you know, they were firefighting legacy code, not a fun place to be. So we had a lot of challenges. One of our challenges really centered around the fact that our entire software architecture paradigm was hardware centric. We didn't want to make large amounts of capital expenditures because it's bad for the business.

So if you look here at the diagrams, you can see that there's a big blue bar that powers everything. And within that bar in our world was our monolithic set of apps and our monolithic database that database did everything for us. It was our reporting database, our operational database, our transactional database, our report processor, our nightly routine runner and we were running out of ways to continue using it because the way that we managed our system was juggling to avoid resource contention.

So we had a job that would run at 8am, 9am and then 8pm, 9pm, 10pm, 11pm. And as the business grew, that job that used to run from 8 to 9, started running from 8:15 to 9:15. And it started running into the next app. And we realized that we really needed to scale up somehow. And we didn't think it made sense four years ago to continue buying hardware and continue scaling up our on premise.

So we decided it was time to make our jump to the cloud. Like many companies that wanna make their jump to the public cloud, we engaged with all the public cloud providers and this is where the AWS team really set themselves apart. They were customer obsessed with us even before we were customers, even before we signed a single piece of paper or spent a penny, they were with us helping us put together a proof of concepts working through some of the challenges that we didn't know how to operate and they brought experts to teach some of my developers on what to do.

So we did what I think many companies do when you have two decades of managing your on premise servers. We're like we have these experts and they're helping us, we can do this right. We're just gonna pick up some servers, we're gonna move them and the magic beans are gonna take us into the promise land.

And so we tried and we took an app on the edge of our infrastructure, we moved it, we lit it up very excitedly and it was a disaster. We had data latency issues, we had concurrency issues. The app was not resilient, it was getting out of sync, it was constantly crashing. We were having trouble getting data up and getting data down.

So we realized we needed help and we reengaged the AWS team and they brought in the partner at our work and of course, Rackspace was one of them and we chose Rackspace because we were optimistic. We were optimistic that this was really the first step of our technical journey and that we were going to continue innovating. This was gonna be more than just a lift and shift.

We wanted to not only lift and shift but throw away some of the old paradigms, try to use bleeding edge technology to help change an archaic talent acquisition industry and in a perfect world, change the world for the better. I mean, how many of you have tried to search for a job and enjoyed it? It's probably not very many. And that's because the entire talent search process really, really, really stinks.

So after we signed with Rackspace, we started the engagement and one of the solution architects pulled me aside at the very beginning and said, "Mark, hear me out, give me 30 seconds before you pass judgment." And he said, "Mark, this is your opportunity to think about this, not just as a technical rebuild, but as a business change to build the business and design your technical architecture in the way that you want the business to be and not the way that your business is today."

So he was right, we engaged with the business teams and we discovered something that many software developers dream of - a lot of business rules that we had trouble wrapping our heads around because they didn't make sense, weren't really business rules or laws at all. They were workarounds created 8 years ago around a technical decision because our old architecture couldn't do it.

So we got the opportunity to throw all that stuff away. We went back to the beginning and we rethought our business, we restructured our domains and we thought of what we believed, a logical way to rethink our entire data strategy should be.

And that question of him pulling me aside, really changed the entire trajectory of the company. So I'd like to ask you guys, is that normal? Is that how you guys engage with customers?

It's funny you say you point me ask a pointed question but I must say one of the things that we always encourage our solution architects, especially when they are the frontline customer facing people, which we always encourage is, preach, do with attitude and do with approach.

So what that primarily means is now work with the customer, empower them on this overall digital journey that they are embarking into. And then if they don't have a buy-in, then I can tell you like the entire project is not going to be successful.

So that's the mantra that we have with all of our solution architects and engineers - work closely with the customer and then more importantly with the business people, it's not just technical people, you need to work closely with the business people as well for the business side, on the customer side as well.

So Mark, when we went down this path together, can you walk us through the architecture, the concepts and how this all came to be?

Yeah, so I still remember when we first walked into JobTarget, like Mark said that one central environment where we had databases running applications running, it was running more like a request response architecture. And especially when you are in this more scalable industry, and I'm assuming and I'm sure like JobTarget has scaled a lot in the last two or three years.

But in fact, double, triple or over 110,000 times, to be honest. One of the primary things in order for any organization to be successful when they are scaling up is adapting modern technologies.

Like Mark said, when the developers started building their application by default, every developer will go through request response architecture. It's good, it works. But when it comes to scaling, it fails.

And a good example is take Amazon.com. Imagine if they are on a request response architecture and there's a Black Friday deal going on or a Prime Day going on. When you make a request, you can't wait for a response for the next transaction to happen. You know, that's exactly the same thing that we have taken to JobTarget and then started preaching them the importance of event driven architecture.

And especially when you are scaling, scaling the organization. And then more importantly, when the organization is moving towards a data driven organization. And that's one thing that I always say to my customers - if your end goal is to become a data driven organization, start adapting event driven architecture.

And when we started designing the event driven architecture for JobTarget, the first thing that we did is tell them what is event driven architecture, how it is important, how it can scale. And then more importantly, how you can grow with that architecture as well.

Yeah, so what you're looking at here is our new architecture. A few slides back, I think I showed you the architecture, we had the big blue box in the middle that powered everything. And what you see on the left is the high level diagram and what you see on the right is a more tactical, actual service based diagram.

But the key difference when you look at this is all the arrows are going in a lot of different directions and this is critical to what we're able to do now and in the future.

So instead of everything going into the main database, when someone clicks an app or applies to a job that goes into that main database and it waits to be processed an hour later or 4 hours later or 8 hours later, that app is gonna get sent into DynamoDB immediately, then it's gonna get picked up by a Dynamo stream, get sent into EventBridge.

It's gonna get emitted into many, many Lambdas where a programmatic app can process it and put it into the campaign in near real time. Real time enough for our customers and then that once that gets into the programmatic app and the customers can use it and understand that they're getting value or if they're really excited about this amazing person that applied to their software developer role, we then take that information and then send it via Glue into our data lake in S3 so that we can continue to use that in our other applications.

This new design required a fundamental shift in how we thought about our apps, we structured our teams and how the entire tech work was really fundamentally organized.

So before when everything was routed around the blue bar, we all made calls against the tables directly. And now if you see many gray boxes, those many gray boxes represent different teams. So when we go back to the business transformation, that's the business transformation.

We went from a team of developers all working against the one thing to smaller teams of developers working around more smaller concrete things, interacting with each other using code contracts, doing it in real time and then having the benefit of infinite scalability using the Lambdas.

If I may ask one thing, Mark, can you talk about your growth, the amount of growth that you had in the last few years and why this thing has helped you in scaling that much growth?

Ironically, the next slide talks about it. So I'm a technologist at heart. I love talking about that stuff, but none of it really matters if you don't change the business.

So this new design increased our capacity by 11,000%. Not only did we increase our capacity by 11,000% we increased revenue by 800%. We did that while paying off technical debt, we did that while retiring old databases and we were able to grow and hire new people and new teams because we had segmented developer teams.

So instead of hiring 40 developers and having to integrate them into this big mass of humanity and have them understand all of our legacy tech and code, I was able to hire 40 developers and split them into 10 different teams. And those 10 different teams have autonomy and authority in their own individual domains that is able to drive innovation for our company and for our customers.

An example of the capacity increase is our programmatic ad builder. So our campaign builder used to take 18 hours to run. It would, it has to manage right around 2.5 million active job postings. And now last night, it ran in 18 minutes. So that 18 minutes gives us a lot more capacity when we go back to - well, now, what can we do?

Well, before we were thinking, what can we cut out? Because what's not adding a ton of value because we don't, we can't run it anymore. The conversation is how much more can we take in? Even if it's a nominal amount of value? And when you just keep piling in more and more signals, more and more information, our processes get better, our models get better, our products get better for our customers.

And the best part, at least to me again, technologies at heart, is with all the data being in S3 and it being more or less available for everyone in the company, we've been able to leverage the most modern technologies.

We've actually been able to bring a new job description builder to market leveraging large language models in AI, helping employers to write more readable job descriptions and helping job seekers better understand what's in their job description.

I mean, how many of you have read a job description and after you read it four times, you still have no idea what the job is? You have no idea if you're qualified for it, you have no idea what your day to day looks like. You have no idea if you can even get this job. That is a myth. And I don't know if we can solve that. It's not magic as much as I want it to be magic. But I mean, it's a step in the right direction.

So Vic, you were a big piece of this. Were there any surprises or aha moments that came along here as well?

I think the growth was something that I wasn't expecting. And I'm just being honest, like 11,000% growth is remarkable.

But, you know, we also need to give credit to the underlying technology that we were able to build for JobTarget. And, and then more importantly, like, you know, all this architecture that we have spoken about even driven and, and job matching and all of that now that has laid the foundation for JobTarget to be more data driven organization.

Now, it is important in the in in current market, like enough for organization to be more data driven. That way you can make decisions based on data. And this is the fundamental like enough foot ground level thing for it to be successful.

Ok. Ok. So you know, moving on uh becoming a data driven organization, it's not an easy thing, but it does pave the path for a whole lot of new opportunities. Can you share maybe you know what you wanna achieve next? Like what's on the, what's next for you?

Yeah. Um so before I go there, I do wanna talk a little bit about what the change in the data driven organization. What are, what has happened in the business?

Um number one, it's been more fun. So instead of me interacting with dull business people, we're asking very basic questions. How are we doing this month and what are we going to do next month? I'm getting intelligent questions. I'm getting questions like our top publisher, our number one publisher is losing share shift to our number three publisher, but only in our top 10% of customers. Do we know? Why do you have a hypothesis as to why that's the case? And those conversations drive immensely bigger outcomes and better outcomes than, than the conversations before where everyone's fighting through like normal processes.

Um but, but uh but looking forward on, on top of that, um we definitely want to keep information security and protect all of the data of all the job seekers who trust us with all of their information. Um the information is very sensitive to them. It's a resume that contains a lot of information about you. So the information security team has a seat at the table and we want to make sure that all of our systems are very resilient and hardened against attack so that we don't lose all of that information and then it's not perfect. Don't get me wrong. It's faster, it's better. We're making more money. All of that sounds great. But uh one of the things that this new design has created is it's created 18 new silos.

So before I had a big blue box and in that blue box, I knew that the data in there was the system of record for our entire world. But now I have 15 teams with 15 purpose built databases with 15 different cadences in which they manage the data. And maybe they even have 15 different versions of the same term. Like maybe someone calls a job description, one thing and someone else puts salary in it. So they call it salary plus job description or something silly like that.

So now we're embarking on a journey to use and build a data mesh thankfully using you guys. Um going back to our original premise that you guys were gonna help us not just do the lift and shift, but to continue to help us innovate. And with that data mesh, I'm hoping that we can bring the 15 silos back together and put humpty dumpty back on the wall, right? Like the problem was, it was big and it wasn't great. Um but it was working but now there's 15 and it's faster and bigger. And now you get into weird data taxonomy issues and data concurrency issues and people having like slightly wrong data which can confuse a customer.

So, and, and the other thing that I want to say, you know, specific to JobTarget and data mesh is, you know, especially when you are growing organization, you want you want data to be available immediately. And if you have a central data warehouse or a central reporting system, and you have to go to a central team to get the data and then you, you do require meaningful decisions out of it. You know, you are losing time and you know, who knows your computer might have another product built already.

So, you know, setting up the data mesh architecture for JobTarget was, was was key for us and it's a key, this is something that we spoke, you know, probably like three months, six months ago to start thinking about the mesh architecture and why it is important how you can have marketing team doing their own data mining and then still rely on on the central architecture to make sure that the governance is in place and make sure they're not misusing the data and everything.

Because there are a bunch of ml models that that JobTarget run and, and if each and every time they come to a central organization, then, you know, it becomes a bottleneck. And that's, that's a key thing that we have.

Um we were trying to avoid, you know, when we is how we design data architecture, once we define architecture, you know, then building the data pipeline and making sure that it's one pipeline that's, that's serving all the purposes and everything and getting the data ingested in right time, you know, right fashion. And then more importantly, the cataloging is happening.

So if, if the cataloging is not happening, then, you know, marketing resource might not know exactly where to go and get it and, and it is, you know, i think that's one area where most of the organizations oversee, they don't do proper cataloging in the meshach texture and that's where they fail. Um so that's another key area um to be to be aware of.

Um and then more importantly, you know, going back to my original comment about if, if you are in this digital data modernization journey, you know, at the moment, you think about even driven, like, you know, my suggestion is by default, you have to think about data mesh. You know, the traditional data browsing is, is is not gonna be helpful there, mark anything to add there.

So one more question question before i move on. Uh so your team is now leveraging the data, they have access to the data. Um you know, what are some of the maybe the comments from the team as, as part of that? Have they they shared any surprises or or aha about that?

Yes, both, both good and bad. Um so, so on, on the good front they like how fast data is there. Whereas before they had to wait until 8 a.m. the next day to start doing their work to see if anything was actually successful.

Um on the bad side, they're impatient and now they see some data coming in, but maybe it's clean and maybe they should use it and maybe they shouldn't use it. Um but thankfully with the business teams and the business partners being more bought in and using some of the more modern analytics tools that we're giving them.

Um the business teams are actually asking these questions to help us manage like some of the taxonomy challenges that vic was, was talking about. I think that's one area that we have seen a job target that i personally have seen in conversations with, with mark and their business team is, is the data taxonomy issue like, you know, everyone understands that, you know, it is there, but no one can confirm. Ok. Is this the right number? And they always go back to my excel says differently, you know, you wouldn't believe how many times we had this conversation around, around excels and we still continue to have that conversation.

But you know, if you, if you want to be a data driven organization, then, you know, let's start moving away from those excels fair enough.

All right. So looking ahead with uh gen a i or maybe a i in general, uh you know, mark, you know, what is uh how is that gonna affect your the future? Do you think? What is, what is, what do you say you see that playing out for a job target?

Ok. So, so for us, um there, there's the most obvious place, of course, you know, optimizing our programmatic advertising system which includes taking things like i was talking about earlier, including trying to process additional signals like whether or not you want to hire remote workers.

Um but there are other products that we're hoping to bring to bear that will help shepherd employers in choosing the right job seekers, solving some things like unintended bias and unintended discrimination and then leverage ways to help screen out things within the resumes.

Um so we, we don't do any resume construction yet. Um but we've already noticed that within the resumes that are coming in, there's been a pretty stark change over the last 8 to 10 months and it's not surprisingly, it's happened mostly on the technical front.

Um where previously you would see very dry, boring resumes from software developers, brilliant people, but they don't communicate well. Um they don't communicate well on topics they like and you expect them to then compose their 20 year illustrious career into a two page resume. And that two page resume doesn't represent how they work. What makes them successful. Should they be in the office three days a week or five days a week? And I'm not sure we can get there any anytime soon.

But I think those are the things that get me excited is maybe we can help, try to find a way and think about, you know, bringing in additional data sources in order to optimize these tools in order to help people get the job that is actually a better fit for them rather than the job that falls in their lap.

Um, I think a lot of people take the job that is most convenient that pays them more money than the job they have today. That's a very normal way to make your career progress, but that might not be the most optimal way to do it.

Um and then I think the the other thing we can do with all this increased capacity is start bringing in additional additional signals that we didn't think mattered quite, quite so much. Um housing build statistics, for example, could, could be a leading indicator as to whether or not there are gonna be more jobs there, more um office emptiness or occupancy rates could also help drive some of these things. And again, with all of the additional capacity, we can start processing all of these things and hopefully optimizing our systems.

Vic how about you? From a Rackspace perspective? What are you, what are you excited about?

So um when it comes to AI or, or generative AI using LLM's, right? You know, from Rackspace perspective, what we have started is you know, we have created a spin off um called FaiR, you know, it stands for Foundry for AI by Rackspace.

Um what we do as part of that is, you know, it's basically a global incubator where we build solutions both for internal use and for external use.

Um the way um I would explain the is, you know, we have four layers or the first layer is the services layer. As you can see, um we have Ideate where we sit down with customers and ideate on a specific use case. And if they don't know what to start with, then we sit down and understand um and define, define a use case that make more sense and, and have a more meaningful impact in the short term. And also long term, once you have an idea, you know, we, we, we take that use case and incubate it and as incubated is nothing but, you know, creating an MVP solution for our customers.

Um you know, pretty much production ready. Um and then the last one as part of our services is in Industrialized, you know, that's where we uh build the application to be more scalable. Um make sure all the governance and controls are in place. And then more importantly, you know, preach responsible AI, you know, responsible AI and ethical AI is, is key these days.

Um you know, um then, then than, than before. Um the second layer that we have is the capabilities. Now, we are a global organization and now we have over 202,000 plus certified skilled Rackers. Most of our Rackers are AI ready in some shape or form.

Um now, that's because we are using for internal and for external as well and we have defined, you know, solution accelerators for our customers.

Um and that's one key differentiator that we do. We have either, you know, a tier three IP or a tier two IP that we have built for our um that we have built and that's a differentiator as part of our FaiR. And then more importantly, our AWS capabilities talks about um about the competences that we have and, and all things included AWS and the third layer as part of our FaiR um incubator program is, is the FaiR AI framework that we bring to the table.

Now, I'll get into um into the details of what AI framework is and how we approach things and all of that. And, and as part of the framework, you know, one of the things that we always preach is like as a responsible AI like anything and everything that we built has to be symbiotic, secure, and more importantly, like, you know, encryption, you know, data is encrypted from end to end.

And then we have um tailor created around 250 plus policies that you have to by default account for when you are implementing a secure landing zone for AI. You know, not many organizations think about it, but you know, it is equally important to make sure that you are, your landing zone is a is is air tight, you know, all guardrails are in place. And that way, you know, you can preach responsibly, you know,

Here is a sample application or a sample architecture that you see, you have a gene application on the front end where users are are interacting with, you know, whether it's a structured data source or an unstructured data source. Uh and all the data in flight is is encrypted. And now if you look at the foundation model, you know, it, it's frozen because once you put it in a secure landing zone, you're not, you're not training that foundation model using, using, using the public data and, and you know, we do not store any customer data or, or parameters as part of a foundation model as well.

So, so those are the key things that you have to account for, you know, when your building is scalable AI application.

Great in the ad mark. No, this is awesome. So we have a recorded video that Vic is going to do a live voice over and we're going to see the place.

Yeah, so one of the thing that I want to showcase here was as part of our FaiR practice or business unit, right? You know, what we have done is, you know, you know, as we work with JobTarget and other customers, you know, what we have built is, you know, we have created a focused application powered by generative AI, you know, the key here is to, we are helping recruiters and interviewers to streamline the interview process. You know, when the key here is responsible unbiased way and everything, we are not asking them to like, you know, make your decisions based on it, but we are helping them to or we are guiding them to make sure that, you know, you have the right questions for the interview and uh interviewer and all of that.

So if you look at this um this demo here, right, you know, we input the type of role that we are hiring for, you know, the company that we are hiring for and the question categories, you know, the question categories is where it's more like, you know, customized from organizations to organizations. You know, in this particular example, our core values are different than, than JobTarget's, core values. So that's where we have to do some engineering and then you input the job description that way.

Um you know, the questions that we are generating using LM are more tailored towards that particular job description. So you input those, those three or four values and then, you know, we generate an interview questions, you know that the recruiter can use through the first level of screening before you pass it on to, to, to um a technical resource.

Um and, and, and the key here is, you know, we are not, we are binary, like, you know, when i say binary, we are not, none of the interviewers are making decision based on this, but we are helping them guide what sort of questions to ask if, if a resume says that, you know, i have worked on even driven architecture, you know, ask them to explain what an even driven architecture is, things like that, that, that it's as simple as that.

So i have two comments on that. One is i think this highlights the evolution of our partnership as it continues to move from migrating to innovating to now. they're helping us id eight.

The, the second part is ii, i remain hopeful with products like these that we are going to keep changing things like um how ii i mean, i, i'm assuming that everyone in the room is technical, maybe that's not true. but when you talk to an hr person who are lovely, they're great human beings.

Um but sometimes they just ask completely the wrong questions and they have no idea they're asking completely the wrong questions. So you go through this first interview with them and it doesn't actually move anywhere and tools like this. And is why i'm so excited about the whole, the whole gen AI from is it's allowing the democratization of technical skills to people who are not historically technical.

And then, you know, i i see it day to day as well. You know, when i'm recruiting resources, right? You know, that data science was, was a good example. If the recruiter is not technical, you know, he can, he can be a good recruiter, you know, go make good connections with the people and bring the right people at the table. But if he's not screening that, doing that first level of screening, then it doesn't help me.

So a product like this or a tool like this, you know, will definitely help recruiters do those initial screenings and then put the right right candidates in front of the leaders.

But all right. So thank you both.

Uh so let's go ahead and close this out. So first, I want to make sure that everyone's aware of the, they can learn more about the two offerings from Rackspace that Vic had mentioned.

Uh we offer an Ideate phase which kind of helps you understand the art of the possible. Uh and then there's an Incubate uh opportunity. It's a very, very rapid aggressive pace to take this uh uh concept and make it an MVP and ultimately industrialize it. And so if you want to learn more, there's some QR codes, it's a really, really streamlined, aggressive uh we think successful uh motion.

And then, you know, the last piece I'll say is if you want to learn more, if you want to talk to us, if you wanna see the image generation that uh Vic has mentioned, come visit us at our booth. We are at booth number 450. Uh we also have a neat VR experience which is pretty cool.

Uh and we'd love to have you, we'd love to talk to you. We love to learn more about your business. What's, what's challenging you and uh share some uh uh some things that we see and things that you see.

So thank you everybody for joining us. Thank you for your time and uh thank you both uh for, for your expertise and for your patience with me asking you questions.