Dr. Xu Miao joined Celential as Vice President of Artificial Intelligence a little over a year ago to bring Celential’s AI-driven technology to the next level. He brings a strong academic background with a PhD in Computer Science from the University of Washington where he contributed research to the fields of Machine Learning, AI, and Computer Vision.
He has a record of success in research advancements at LinkedIn and Microsoft, as well as leadership experience in AI at multiple startups.
We chatted with Xu about the path that brought him to Celential, including his initial encounter with Celential’s Virtual Recruiter Service. From there, we discussed Celential’s highly vertical, “white box” approach to AI, and how the collaboration between human experts and AI can accelerate the development of AI solutions to recruiting problems.
What drew you to computer science as a field in the first place, and what led you to pursue AI in your graduate work?
My undergrad actually wasn’t in the field of computer science at all; I was an electrical engineering major. It is a mixture of physics, hardware, and a little bit of software.
But after graduation, I figured out I was not a hardware person. I can hardly assemble a radio. So I decided that I’d pursue a career in a software or a mathematically-oriented field. That’s where I started to do computer science work.
The mathematics for AI are quite similar to the mathematics for the physics on the electrical engineering side, and it was easier for me to pick up the theory and the practice for AI.
That’s where I started to feel like “Oh, I’m connecting with this,” and I just went down the rabbit hole for many years.
There are two major shifts that you’ve undergone as a professional. First, you went from being in academia doing visiting scholarships and postdoctoral work to working in industry at larger organizations like Microsoft, LinkedIn. Then you went from larger companies to being a co-founder and now a VP in a Series A startup. Could you tell us about what inspired both of these shifts? What drew you from one place to the other, and what do you find similar and different about them?
In general, I think all three environments are very challenging, and while they’re very different, they share one thing: they want to innovate. I’ve always been motivated by innovation, so I’ve pursued innovation in different forms throughout my career.
After I got my PhD, I was waiting for opportunities in academic positions. But innovation is everywhere. It’s not just in academia.
I found out that Microsoft opened up a secret project on computer vision using the second generation Kinect camera. It was part of their research branch’s incubator program that pushes research work into real products.
That’s the part that immediately resonated with me. So I joined Microsoft to follow this path.
After a couple years in a large company, I realized that if you really want to create something completely new, you better go to a startup. That’s not to say that big companies don’t create new things. But they rarely take on a project that’s less justified for the value.
They act more as a steward for ideas where you just need to figure out the technical part.
If you want to go to the front line and test something out, then that’s the realm of startups. A lot of things–even with, say, deep learning–come from startups. Like DeepMind–they beat that Go master with AlphaGo and then they got acquired by Google.
That’s how startups contribute to innovation in the industry. And that fascinated me for a while, and I started my journey in startups and I haven’t turned back yet.
Let’s talk a little about your experience encountering and then later joining Celential. What was that initial outreach from Celential like? What eventually led you to sign on with Celential?
It’s an interesting story. I found out about Celential when I was contacted by their CEO, Andrew Dong, through their AI system about a position for one of their early customers.
I was impressed by the personalized email for a couple reasons. He spoke about my prior work, and pitched the idea of building something that could change people’s lives. The outreach was quite touching compared to the more bland emails recruiters often send.
I immediately replied to Andrew to set up a meeting.
But the timing wasn’t right to leave my role at the time, so Andrew and I became friends afterward instead. He even became an advisor for my company on business interactions and product direction for a couple of years.
That role didn’t ultimately pan out, and I felt like I needed to find something new. So I talked to the team at Celential. I felt certain that it was a really good place. There were a lot of smart people who were very dedicated to fulfilling their mission. I got along really well with the team in the interview, so I decided to join.
The first version of Celential’s talent graph and virtual recruiter product had been developed by the founding team by the time you joined, but since then you’ve worked to level up these products significantly. What previous experiences did you draw on in this work?
Well, most AI companies are building systems that are built as ‘black-boxes’. The GPT-3 model by Open AI is a well-known example. They don’t need a domain expert to check it at all. They just feed massive data into the AI algorithm and extract interesting outputs based on correlations of data features.
They have shown marvelous achievements on open domain problems, but still quite far from providing solutions in deep verticals or specific challenges because it’s difficult to evaluate or to speed up the machine learning and improve the technology so that it can still fit into the vertical.
I believe the fastest way to adapt the latest AI models to a vertical domain is through the ‘white-box’ or ‘open-box’ AI technology. That was what I was developing before joining Celential.
That approach seeks the answers to problems the black box approach doesn’t concern itself with. Problems like how you explain what the AI’s actually doing, or how you encode domain knowledge into the AI algorithm, or how you get the AI model to reason more like a human with domain expertise.
And that’s the reason I wanted to join Celential: I believed I could bring that value to their talent graph. That’s our focus of the data team — enabling human experts to train the AI efficiently and effectively by utilizing their domain knowledge.
For example, to source software engineers, you need an incredible volume of specific data about their skills and professional trajectory. This specialized data can come from professional communities, personal websites, company profiles, and places where software engineers talk to each other online.
Celential was already able to capture this data via the talent graph before I joined, so my role is to feed our AI that data and tune the model as if it were a high performance car engine.
With this engine running on that specialized data and our engineering team tuning it to reach peak performance, we developed an AI that can think proactively about people’s career paths.
Our AI can even predict the likelihood that a strong senior software engineer or tech lead would be ready to be an engineering manager in their next job, or how likely someone will want to move from a larger to a smaller company or vice versa in their next role.
It will then match that person to an open role from one of our customers.
These are the kinds of advancements we can make by going deep into our vertical, and they’re often the situations that expert recruiters face everyday but can’t perform at scale.
With that vertical, white box approach to AI in mind, what is Celential’s unique value add as a recruiting solution? What is Celential bringing to the market that something like LinkedIn might not be bringing?
I think that LinkedIn Recruiter has a different direction. They’re trying to address something broader, trying to reach a broader market rather than focus on verticals.
They’re trying to say “Okay, we’re going to take on all recruiting problems, we’re going to take on all kinds of roles, from the technical, technology-driven roles to, say, chefs.”
And that makes the AI solution or algorithm harder to refine because it’s so broad. You can only scratch the surface, and in the end the candidate matching quality cannot be guaranteed. So the hiring company has to use a human recruiter to use their solutions.
Although it does help the recruiter to get better candidates, it’s not a fully automated process. You still need a lot of human involvement in that process.
But our virtual recruiter solution takes a different route. We don’t want to address the broader market, we want to address the software development vertical.
That’s the mission that we have. Because this is a vertical domain, I believe the collaboration between domain experts and the AI algorithm will speed up the development of the AI system.
At the end of the day, I think we can create an end-to-end automated solution and user experience for the hiring company, one that has both a lower cost per hire and a more streamlined process for hiring.
Interested in learning more about how Celential is solving companies’ tech candidate sourcing challenges with Artificial intelligence?
Schedule a demo today!
Table of Contents