Mastering the Art of the Interview: An AI-Powered Approach
Interview with Kevin Hsieh, founder of Mirwork.ai
Kevin Hsieh is the founder of MirWork.ai. An AI-powered mock interview tool to help tech. professionals find their dream job.
Before this startup, Kevin held senior product leadership roles at Meta and Cognoa, where he led significant projects like Reality Labs and AI-driven diagnostic tools. He has over 13 years of product experience, known for launching 14 innovative products across sectors like IoT, robotics, and medical electronics.
In today’s interview, we talk about:
MirWork’s mission
Why mirwork.ai
Long term vision
Challenges with building AI products
Kevin, good to finally have you. Let's start from the top. Can you please share with our audience what problem you are trying to solve with MirWork.ai?
Kevin Hsieh: Mustafa thanks for having me.
So the problem I'm trying to solve is helping people practice for interviews by using AI. That's why I started my company MirWork - it stands for "mock interview robot" because it acts as that practice mirror.
See, when I was prepping for my final interviews at Meta, doing mock interviews with other PMs was game-changing. It allowed me to understand my capabilities - not only my strengths but also areas I needed to improve. That deliberate practice is what gave me the confidence to represent myself well during the actual interviews. And I was lucky enough to get the offer!
I feel like everyone should have access to that level of productive practice. So after joining Meta, I started my small PM community of just a few people I could mock interview with. It's grown to around 100 now.
That is a great story. Sounds like you were already doing this before AI. When did you start using AI in your product?
KH: Pretty early on. When large language models came around, I saw the potential to use AI as a tool to facilitate that continuous mock interview practice in a more scalable way. The truth is, when I was tutoring around 80 students over the past couple of years, I didn't have enough bandwidth to provide truly personalized coaching and tracking for each person's learning journey.
By incorporating AI, we could ensure their progress and areas for improvement are reflected through the mock interviews. That's why MirWork was born - to be an awesome resource and guide for any tech professionals looking to become the best version of themselves through productive AI-powered practice. It started as just a small idea with some friends from the community, and we're extremely excited about the potential!
May I ask, where are you in your product development lifecycle?
KH: We have our first MVP out. And this one is focused on helping PMs with building “product sense”. It’s a way of thinking about products. And I started here because this is where a lot of people failed during their interview with Meta.
Plus I am a big fan of product sense. In some ways, I use it even when I buy tech products for my friends and family.
So do you mock out the whole interview?
KH: Yes. For example, if you are applying to Google. The AI will conduct the entire product sense interview with you. It will ask you PM questions like - how would you build a refrigerator for the blind? Or what would you do if you were Google’s CEO for a certain interview?
Plus, my favorite part, at the end of the interview, is that it will give you feedback on what you did well and where you can improve. And all of this is based on my personal experience plus with coaching a ton of folks on interviews.
That’s great. May I ask, where do you want to take this the next year? What new capabilities are you thinking about?
KH: Yeah, there's a ton on my mind for MirWork's roadmap! The product senses mock interview is just the MVP. From here, I want to focus on three things.
More customization to better support candidates. Soon, you'll be able to upload your resume and a job description you're interested in, and the AI will tailor the mock interview to your specific needs, timeline, and situation.
Build connections with the tech community. For a long time, I've been helping folks network - like if someone reaches out saying "Hey Kevin, I'm interviewing at Google Cloud in 2 weeks, do you know anyone I can practice with there?", I can quickly match them with people I've mock interviewed before that they can trust. I want to build that same capability into my product.
Explore new capabilities. It's incredibly exciting to think about where AI capabilities could go next beyond just conversations. Things like capturing video or voice sentiment could unlock whole new interactive experiences. Maybe we won't even need full dialogs, and AI can almost read your mind with just keywords to surface the perfect coaching. That’s one of the topics on my long-term vision.
It's both scary and thrilling to imagine those future possibilities sometimes. For now, I'm just focused on iterating our initial mock interview experience rapidly.
One last question, as you are building your product. What has been your biggest challenge so far?
KH: To be honest, there are so many (laughing).
There are three that come to mind.
First, the rapid evolution in the AI landscape. Since these large language models emerged, it feels like every week there's a new model, technique, or company making waves. As a startup, even if you have a solid foundation, it's hard to navigate that noise and figure out the optimal path forward quickly.
Second, is prioritization. There are so many ideas and features we want to build, but which ones do you prioritize for your specific use case? How do you even define the criteria? It can be overwhelming for our lean engineering team at times.
The third is just GTM. How do we get our word out? I have my network, but if this is going to grow, we need to go beyond my network.
Have hallucinations been an issue as you are building your product?
KH: Hypothetical interview scenarios work in our favor. Given our unique focus on mock interviews. It does allow a bit more flexibility around things like hallucination. Since we're facilitating hypothetical conversations, the interaction between students and AI is more important than perfect factual accuracy in certain areas.
At least for now, I've been experimenting with techniques like RAG and unsupervised fine-tuning to find the right balance. Using it as an interview practice tool fits nicely. Data privacy is still top-of-mind, but I'm less worried about factual inconsistencies for this application.
Kevin, this has been great. Thanks for being a guest.
KH: Mustafa thanks for having me. It was great connecting with you. And this was a lot of fun. And if any of your readers want to try out MirWork, please visit our website.
Thanks for having me!