Founder Interviews: James Wu and Allen Lu of Adaptilab

AdaptiLab cofounders James-CEO (left) and Allen-CTO (right)

Childhood friends James and Allen are making deep learning easier to master with Adaptilab.

Davis Baer: What’s your background, and what are you working on?

James: Hi I’m James. I’m the CEO of AdaptiLab. I recently graduated from Duke University, where I double majored in Computer Science and Statistics, and I’m currently a Duke Melissa & Doug Entrepreneur. I’ve worked at two startups over the last three years before cofounding AdaptiLab, and I was a 2017 Freestyle Venture Capital Engineering Fellow (worked at AdStage).

Allen: Hi I’m Allen. I am the CTO of AdaptiLab and a recent CMU graduate. I studied Computer Science with minors in Machine Learning and Language Technologies. I worked on deep learning teams at Google and Microsoft before cofounding AdaptiLab, and I am a published deep learning researcher with my work on spoken dialog systems and text summarization.

James: We both grew up in Seattle, and we’ve been friends and neighbors for the last 11 years. We both graduated a year early from university to move back to Seattle to cofound AdaptiLab in May 2018. AdaptiLab builds interactive training modules, “labs”, that teach engineers the practical skills and intuition to train deep learning models. The labs are presented in an integrated development environment (IDE) directly on the website, so users receive their instructions and write code all on the same screen. Each lab covers a different real-world industry application (e.g: image recognition, data preprocessing). After completing the Labs, engineers will be able to build an end to end machine learning solution from preprocessing input data to training and tuning their coded models and interpreting results. AdaptiLab launched on September 15th and we have reached $1000 in revenue in individual lab sales and are in discussions with several companies for selling enterprise training packages to engineering teams.

What motivated you to get started with your company?

Allen: We have deep domain expertise in machine learning, both in industry and academia. We witnessed firsthand how poor training solutions are in today’s market, even at the most successful tech companies. Interns and software engineers often have to learn practical deep learning on their own on the job by following online tutorials, reading documentation, and copying examples in github repos. This learning strategy is slow and ineffective. Companies must wait for engineers to learn on their own, or the company needs to train the engineers themselves. Furthermore, none of these “self-learning” solutions teach the intuition you gain from years of experience tuning and evaluating neural networks on complex industry problems.

James: Modern solutions for teaching deep learning engineering focus on theory. University classes and online courses taught by Professors, such as those offered by popular MOOCs, teach material as if they’re preparing students to go do graduate school research. However, most students are really just interested in being able to code useful and interesting deep learning applications on their own or for their job.

We believe strongly in the interactive approach to teaching engineering. Codecademy proved this thesis by teaching web development with an interactive IDE experience. We ran a private beta with testers from both industry (engineers, data scientists, consultants) and academia (students, researchers), and we received feedback that the interactive IDE experience teaches deep learning engineering faster than online classes because each lab is essentially a deep learning project that students code all on their own.

Video-based classes teach theory and then assign code to supplement the lecture, so the assignments are unstructured and often unrelated to each other, which is the exact opposite of an industry engineering project. Interactive classes teach code and then provide instructions and explanations to supplement that code, so the instruction style is more oriented towards building end-to-end solutions for industry applications and problems.

Allen: Both of us are published machine learning researchers and experienced engineers (combined 5+ yrs of experience). We were also both teaching assistants in college. Our backgrounds gave us the perfect tools to solve this problem.

What went into building the initial product?

James: AdaptiLab officially launched after four months of development of the product, but we built several platform versions beforehand. After cofounding the company, our first step was to validate the effectiveness of our interactive training platform with potential users. We wrote our first Intro Lab, which offers 5 hours of coding instruction and introduces the learner to neural networks, and then built the first interactive IDE on the site. We deployed the first version of the site 3 weeks into founding AdaptiLab and immediately began acquiring beta testers to test the Intro Lab content and the interactive learning experience. The response was overwhelmingly positive towards both our lab content and the teaching platform.

One of the early designs for our online IDE

Allen: Following a successful beta test of the Intro Lab, we spent the next two months developing our first two “Core” Labs: Data Lab, which teaches data preprocessing to prepare data for input into deep learning solutions, and Image Lab, which teaches several industry standard models for image recognition and tackles a couple famous Image recognition datasets. These labs total about 40–50 hours of coding instruction each. We completed both labs around August 17th and began testing the content with our beta testers.

During this development time we were also working with a designer to build a complete design of the website and the user experience. We switched to building the complete website after launching the first three labs, and officially launched on September 15th.

The completed IDE experience at Launch

James: Having just graduated, we were lucky to not have any financial burdens or responsibilities to address before cofounding AdaptiLab. We are both working fulltime on the company without any foreseeable distractions in the future. Stripe and AWS are both very supportive of startups and we were able to use Stripe Atlas to incorporate as a Delaware C-corp relatively cheaply. We were awarded many free AWS credits for hosting the site initially. We also received initial funding from my position as a Duke Melissa & Doug Entrepreneur that helped us get the company rolling smoothly, and now that we have launched, we plan to raise more funding and to move down to Silicon Valley this winter.

How have you attracted users and grown your company?

James: Following our launch, we have mainly attracted users via word of mouth, content marketing, and social media. We were able to email blast the launch to both of our universities alumni and student mailing groups. We maintain a company blog where we discuss our industry, the future plans for AdaptiLab, and our own backgrounds. We also have a LinkedIn to announce important events in the company, such as our launch. We decided on LinkedIn as the most appropriate form of social media for marketing because AdaptiLab is geared towards industry training. Finally, we both reached out to our many industry and academia connections to facilitate our initial enterprise sales efforts. For our individual sales we chose relatively simple, yet effective marketing techniques, getting the news of our launch out to as many people as possible. The Enterprise sales cycle is a much longer process and we were thankful to have a lot of helpful connections that were able to get us leads with companies.

We believe at the young stages of a startup (pre-revenue or small revenue) it does not make sense to pursue paid growth channels, such as advertisements, because the initial stages of a launch should be focused on finding product market fit and refining the product with feedback from users.

Allen: We setup metrics tracking systems to evaluate how users interact with the site. We were especially concerned about the onboarding process to the learning platform because that would often make-or-break whether a user decides to register an account. We initially used Google Analytics (it is very easy to setup) to track user metrics, but we found that it did not provide the most clear representation of the user experience. We’ll be integrating Mixpanel soon to improve our metrics tracking.

What are your goals for the future?

James: We want AdaptiLab to become the industry standard for training deep learning engineers. We think we have a lot of opportunities to expand into recruiting and competitions, but first we’re going to build the best possible online learning experience for an engineer to build the skills they need to complete industry deep learning projects. This includes completing our “Core” curriculum of labs-5 labs that will provide a robust overview of deep learning engineering-and expanding into our “Advanced” labs, which will cover the hottest new discoveries in deep learning, such as data generation and adversarial training.

Allen: Artificial Intelligence is one of the hottest fields in tech right now. People are coming up with new applications everyday across industries-healthcare, self-driving cars, voice assistants, intelligent robots-and deep learning algorithms are driving all these new innovations. As the world continues to come up with these exciting applications of deep learning, the need for competent deep learning engineers will be greater than ever.

James: We think our major challenge will be proving our effectiveness against other deep learning education courses out there. However, none of the other courses focus on bridging the gap between the theory taught in classes and the practical coding skills needed in industry. Once we demonstrate to our market that we not only provide the background in theory, but also teach the necessary skills and intuition to build end-to-end deep learning solutions, we strongly believe companies and individual students will recognize AdaptiLab as the go-to for deep learning training.

Have you found anything particularly helpful or advantageous?

James: I think we have some mutual character traits that have been very helpful in building AdaptiLab quickly. We are both open-minded and flexible when it comes to trying new ideas and pivoting away from old ones that prove to be ineffective. This is crucial for a startup because so much of building a successful product is the process of building quickly, getting feedback from users, and then iterating on the product.

Allen: We’re also both organized. When first starting a company, there are a lot of logistics issues that need to be taken care of even though the main focus should be building the product and then selling it. We budget our time well to make sure we cover all the necessary legal and finance tasks, without letting them distract us from the main objective, which should always be building the product and selling it.

Because we graduated recently we still have access to many of the connections we built during our time in university. That has been a big help for our initial sales. Most startup founders at best only utilize their university’s alumni network, but we’ve also been able to seek advice and assistance from former professors and research advisers as well as talk to students directly.

What’s your advice for entrepreneurs who are just starting out?

James: There are so many great resources out there for new entrepreneurs to build a company. You should try to read and learn as much as you can and then create your own thesis for how you will run your company. Many of these resources are completely open-source as well. YCombinator’s startup school lectures and conversations have been particularly helpful.

Many universities have Entrepreneurship pre-accelerators and they’re definitely worthwhile to check out if you’re a student founder. My position as a Melissa & Doug Entrepreneur at Duke gave us an awesome group of advisers and a network of connections to reach out to for advice or even sales.

Try your best to find a co-founder and make sure at least one member of the founding team is a technical co-founder. Having a technical co-founder on the team enables the company to build and test ideas quickly. A co-founder does more than help build your company with you, they’re also there to share the entire experience. The founding team will learn together, share hardships, and constantly improve each other and the company.

Pick a co-founder that you trust and that you know has a similar drive and passion. Allen and I have been building products together for over 5 years and we’ve known each other for more than double that, so we can build on each others strengths and communicate effectively.

Where can we go to learn more?

Check out our product at You can also follow our Linkedin for news and our blog for more about the company and the founders.

We’re always happy to hear feedback from users and answer any questions regarding the company. Please feel free to contact us here. If you would like to connect with the cofounders, you can find us on Linkedin as well: James, Allen.

read original article here