Sakana AI Software Engineer Interview Process

Sakana AI isn't a name most people outside Japan know yet, but in the AI research world, they're turning heads. Founded in Tokyo by David Ha (former Google Brain) and Llion Jones (one of the co-authors of the original Transformer paper), they're building nature-inspired AI  systems focused on smaller, more efficient models. When a recruiter reached out, I jumped at the chance.

Recruiter Call

The first step was a 30-minute call with a recruiter. Standard stuff: my background, why Sakana, what I'm looking for. They asked about my experience with AI/ML infrastructure and fullstack development. Sakana is a research lab, but their engineering roles require broad technical skills since you're building the infrastructure that powers the research.

The recruiter mentioned that Sakana values "understanding over implementation." I didn't fully grasp what that meant until later.

Technical Assessment (Take-Home)

Here's where Sakana diverges from everyone else. Instead of a timed coding test or a live coding session, they gave me a take-home project with an extended deadline. Not a weekend. Not 48 hours. Up to one month.

At first I thought this was generous. Then I realized it was a trap of sorts. With that much time, there's no excuse for a shallow solution. They're not testing whether you can solve a problem under time pressure. They're testing whether you truly understand what you build.

I spent about two weeks on it, going through multiple iterations. The key thing I did right: I documented everything. My README explained my hypothesis for each major design decision, what alternatives I considered, what limitations my solution had, and what I'd improve given more time.

When I talked to someone who had done the same process, they told me they spent three weeks and submitted a polished project but bombed the follow-up interview because they couldn't explain why they made certain choices. The code was good, but the understanding was missing.

The lesson is clear: Sakana cares more about your decision-making process than your final output. A simpler solution with well-reasoned decisions beats an over-engineered one you can't fully explain.

Technical Interview

This was a 60-minute experience deep-dive. No whiteboard. No LeetCode. The interviewer pulled projects from my resume and went deep.

They picked one of my past projects and asked me to walk through it end-to-end. Then the follow-up questions started. "Why did you choose this database over that one?" "What was the hardest bug you encountered?" "What would you do differently now?" Every time I gave a surface-level answer, they pushed for specifics. Numbers, metrics, exact technical details.

They also tested fullstack knowledge. Even though I'd mostly worked on backend systems, they asked about frontend considerations, deployment strategies, and infrastructure choices. Sakana builds research infrastructure that spans the entire stack, so they want engineers who can think broadly.

The most interesting question was: "Tell me about a technical decision you made that you later realized was wrong." They wanted to see intellectual honesty and the ability to learn from mistakes. I talked about a caching strategy that seemed clever at the time but created consistency issues we didn't catch until production. The interviewer nodded approvingly when I explained how the experience changed my approach to caching in subsequent projects.

System Design (Optional)

I have heard system design round but it depends on the team and your interview performance. I will update this when I heard back more regarding their system design

Culture Fit

The final round was a 45-60 minute conversation about how I work with people. Sakana is a startup that works with enterprise clients, so they need engineers who can communicate with non-technical stakeholders.

They asked about times I navigated ambiguous requirements, translated business needs into technical solutions, and collaborated across
teams. One question that stuck with me: "Tell me about a time a customer or stakeholder asked for something technically infeasible. How did you handle it?"

They're clearly looking for engineers who can operate in the messy space between research and product, where requirements change fast and you need good judgment about what to build and what to push back on.

Summary

Sakana AI's interview process is refreshingly different. The month-long take-home project rewards depth over speed. The technical interview is a conversation about real experience, not a coding exam. And the culture fit round tests whether you can operate in a fast-moving research startup.

The biggest risk is underestimating the take-home. You have time, which means they have high expectations. Document your thinking, acknowledge limitations, and be ready to explain every decision you made.

If you're the kind of engineer who geeks out over understanding why things work (not just making them work), Sakana's process will feel natural. If you prefer the structured predictability of LeetCode rounds, this will be uncomfortable in a good way.

The whole process took about five to six weeks, partly because of the extended take-home timeline. Communication was responsive throughout.

For anyone interested in working at the intersection of AI research and engineering in Tokyo, Sakana is one of the most exciting opportunities right now. The founders' pedigree is world-class, and the problems are genuinely interesting.

Want to save your preparation time? Check out https://furustack.com to understand what you need to prepare instead of guessing.