In December 2018, after finishing my PhD in Computational Linguistics in Germany, I landed in Tokyo to start a new chapter at an international research institute. The academic life was fulfilling. But as I hit my three-year mark, I started getting curious about the industrial side of things, particularly the prospect of generating real-world impact through products. Motivated by this, I began to explore the path of transitioning to industry.
After a few months of preparing and interviewing, in May 2022, I made the leap, joining SmartNews as a machine learning researcher, where I dove into the fast-paced and hands-on realm of industrial research in natural language processing (NLP) and news recommendation. This transition was a major shift in my career, offering new challenges and learning opportunities. After a year, I took another step forward, advancing to a senior researcher position at Money Forward Lab. Here, my work centers on large language models (LLMs), such as hallucination detection in LLMs. It’s been quite an adventure, and I’m here to share the experience with you.
Navigating the transition
Understanding the landscape
After deciding to move to industry, I started to learn what opportunities are out there by contacting recruiters and talking with friends who are already in industry. After several discussions, I learned that, in industry, hands-on experience is more valuable than research backgrounds. Moreover, the interview processes often include assessments on technical aspects such as system design, about which I knew nothing at that time. It became clear to me that I needed to broaden my technical skillset and sharpen my interviewing skills first.
Expanding technical horizons
Although the tools and technologies in demand can vary in different companies, I found that a solid understanding of software engineering (e.g., system design, Kubernetes, Docker, AWS) and familiarity with popular machine learning toolkits (e.g., Pytorch, Tensorflow) are essential to most machine learning scientist/engineer roles. While I’m confident in my knowledge of machine learning and NLP, software engineering skills, which are cornerstones in many industrial applications, are less emphasized in academic settings. After identifying what specific skills I needed to learn, I dedicated considerable time to catching up.
While working full-time, catch-up preparation can be daunting and lonely. Fortunately, a good friend supported me throughout the journey. My friend helped me divide my preparation into several sections: coding, system design, machine learning system design and technical question response. I made study plans to balance daily work with interview preparation, and I particularly enjoyed preparing for system design part. I learned the intricacies of modern software, like choosing the right databases and understanding how components like load balancers and microservices contribute to a system’s efficiency. The preparation was really rewarding. The knowledge not only helped me land a job, but also prepared me to collaborate effectively with colleagues from infrastructure, frontend, and backend teams in my new role.
Sharpening interviewing skills
Knowing the details of essential skillsets does not guarantee good performance in interviews. Clearly explaining technical ideas and positioning oneself as a competent coworker requires practice, especially when being under pressure in interviews. During the preparation stage, I had several mock interviews with my friend to assess my progress and get feedback. After each mock interview, I reflected on my performance and tried to find ways to improve. Several weeks into my preparation, a recruiter from SmartNews reached out to me. I was not fully confident, but decided to take the challenge anyway. And the rest is history.
Embracing new cultures and challenges
Before my transition, my friend in industry had already highlighted some key cultural differences between academia and industry. Despite being well-prepared, I still faced significant challenges, as industry work comes with its share of frustrations.
Fast-paced development cycle
First of all, I was truly amazed by the fast-paced development cycle and the importance of incremental delivery in industry. It took me conscious effort to adapt from a perfectionist research background to the fast-paced development culture. For example, instead of working towards a perfect solution, I learned to quickly test ideas with rough demos. Initially, I was so scared of delivering less-than-perfect machine learning models. To overcome this mental barrier, I planned follow-up improvements to avoid potential embarrassment. To my relief, my coworkers prioritize progress over perfection. They never blamed me for delivering a less-than-perfect model. Instead, they cheered me on for delivering on time. Over time, I have not only grown more comfortable with the fast-paced development culture, but have also assisted a new team member in breaking down her project into several modules with milestones to ensure speedy delivery.
Effective communication
Moving from academia to industry also requires adapting to a new communication style, a change I initially wasn’t fully prepared for. In academia, I was surrounded by fellow researchers who also work in NLP, allowing for effortless work-related discussions without concern for mutual understanding. However, in industry, I meet people who tackle different tasks and are with different backgrounds and interests. This poses a significant challenge in communicating effectively.
Interacting with product managers taught me valuable lessons. As a researcher, I tend to focus on the specifics of my machine learning models. In contrast, product managers focus more on prioritizing a project’s potential impact, execution timeline, and required resources. Initially, I found it challenging to launch any project. It was frustrating. With reflections and coaching from my manager, I gradually learned to adopt PMs’ perspectives and proactively address their concerns with evidence. Launching projects became much smoother.
Constructive mindset
Unlike independent projects in academia, success in industry heavily relies on teamwork. Projects may halt due to strategic shifts or lack of resources from other teams, leading to stress, conflicts, and frustrations. Through these experiences, I’ve learned the importance of navigating frustrations with a constructive mindset to foster effective teamwork, such as maintaining respect in the workplace, always seeking ways to contribute, and making it a point to celebrate my colleagues’ achievements, no matter how small they may be.
For example, I learned that sharing learnings with colleagues is an excellent way to foster collaboration. Once, a product manager shared his slides from a project I wasn’t involved in. After reading the slides, I thought a paper I had recently read could be relevant. So I shared the link in the comment section. It caught his interest, and this sparked a new project involving 3 teams. We quickly developed a demo of the project for the company’s annual Hackathon. And we won the Grand Prize!
Looking back, moving forward
Looking back, the transition journey was full of challenges, but also rewarding. I learned a lot and grew a lot. Having gained experience in both worlds, I now feel better equipped to tackle new challenges. However, if given the opportunity to go through the transition again, I would spend less time on self-doubting and worrying. Instead, I would prioritize actively seeking information and acquiring new skills.
Doubts about skills and experience are common when making a career change. However, with available resources, such as ChatGPT, LinkedIn Learning, and Coursera, one can rapidly build any required skill. Moreover, the tech community is often open to welcome new talents. So, if you are also considering a career change, fear not. Stay informed about the essential skills for your dream job, continue developing new skills, and showcase your learning through projects or experience. Dive in and embrace the change!