It All Started with an Interview Round I Couldn't Skip

Translated from the Chinese original · Read the original

It started with a low-ROI task I couldn't skip

I've been preparing for interviews lately. Of all the rounds, the coding interview is the most basic one. Whatever your level, if the role involves building software, almost every company will give you one.

But if you've been working for many years, especially as you get more senior, your day job looks very different. Whether you're an IC or a manager, most of your time goes into high-level product, technical and design decisions. When you do write code, it's mostly business logic. You implement complex algorithms less and less, and when you do, nobody asks you to finish one in a few minutes under pressure. The interview asks for exactly that: solve an algorithm problem in 10 to 15 minutes, under pressure, with no bugs. Getting yourself back into that state often takes more time than preparing for any other round.

Put plainly, the return on that time is low. Still, if you want to pass the interview, it's a hurdle everyone has to clear.

What actually makes it hard?

After preparing for a while, I noticed I kept getting stuck in three places.

The first was forgetting. I didn't have much time to practice, so my preparation stretched over a long period. Problems I had once solved fluently felt unfamiliar when I came back to them a few weeks later.

The second was that the way I practiced didn't match what interviews evaluate. My usual routine on LeetCode was to read the problem, run through an approach in my head, start coding right away, test with the examples in the problem, then fix the code or submit based on the results. Once it passed, I counted the problem as done. If I wanted to review it later, it was hard to find my old attempts or remember how well I'd actually done. An interview looks at other things. First, can you explain your approach clearly to the interviewer? Note that this means saying it out loud, not thinking it through in your head. Second, can you write the code in 10 to 15 minutes, design your own test cases, dry-run them, and debug on your own when something goes wrong? When you practice alone, it's hard to make yourself check all of these every time.

The third, and the hardest: after solving a fair number of problems, I still didn't feel sure I was ready. Was I improving? By how much? I had no way to measure my progress.

At work, I've always said that you can't optimize what you can't measure, so before choosing a solution, the first thing to pin down is how you'll measure it. Yet when it came to my own interview prep, I couldn't even say where I stood. That was something I could not accept.

My first user

I remembered something my mentor told me in my first year at work: if you find yourself doing the same inefficient thing over and over, think about building a solution that fixes the inefficient part systematically. That has stayed with me as a working habit and a way of thinking ever since.

So I built a tool, and I was its first user. Before I started, I did look at what was already out there. If a ready-made solution existed, I would just use it. Unfortunately, the existing coding interview prep sites and tools focused on problem lists and solutions: company-specific lists, "master algorithm interviews in a month", and so on. They sound appealing. But is that what you need most while preparing? Are those really what stand between you and passing the coding round?

If the market doesn't have the product you need, you might as well build it yourself, especially now that AI tools are this good. As someone who has been building products for years, my first reaction was: enough talk, just build it.

The first version had only three features, one for each of the three challenges above:

  • Log every LeetCode submission along with notes on the mistakes I made, and rate myself on the same dimensions a real interview uses, so my practice targets what interviews actually test
  • Schedule reviews at regular intervals, to deal with forgetting
  • Visualize my progress, so I can see exactly where I stand

I shared it with a friend. He said entering every submission by hand was a hassle. I felt the same way. So I looked into how Chrome extensions work and tried injecting a button into the LeetCode problem page. One click saves the problem's metadata and my latest submission to my tool, which removes the most time-consuming part: manual entry.

At that point, it had become a tool I wanted to open every day, and it really did make my review and preparation more efficient.

Helping more people

My mentor's advice actually had a second half: when you build a solution for yourself, ask whether it could help more people too.

Almost every friend of mine who wanted to change jobs felt intimidated by coding interviews. When we talked, they were stuck in roughly the same places I was, and many of them had given up because of these same problems. But most didn't realize that these problems were why they couldn't keep going. They kept spending hours hunting for interview experience posts and problem lists, as if those were the cure. Once I pointed out the real reasons they kept stalling, it clicked for them. So I decided to open the site to everyone, hoping to help people who felt lost prepare more efficiently. That's how CodingPartner came about.

Later, I wrote a post in the LeetCode community about the pain points I ran into while preparing. It got more than 300 views overnight and passed 1,000 in two days, and it stayed at number one on the trending list for a full week. Many people said in the comments that these were their pain points too.

I didn't expect that kind of response. The post brought the site its first users. I didn't spend my energy seeking validation in the comments, or letting the criticism of my writing style get to me. In the spirit of experimentation, I treated the comment section as free user research and looked through it for directions the product should take next.

Once I had users, I noticed something on the user-behavior dashboard I maintain:

Quite a few users were already logging problems and reviewing on the site, but all of them were entering problems by hand.

The extension could do that in one click. Why weren't they using it?

My guess was that they probably hadn't found the extension. At the time, it only appeared on the landing page and in the navigation bar. The first page users saw after logging in had no introduction to it and no way to install it.

When I saw this data, my first idea was to add an extension entry point to that first page after login. I tried it myself and it felt out of place. After logging in, the user's next step isn't necessarily to log a problem, and a sudden extension prompt there made the page feel worse. (Yes, I care a lot about whether a product feels good visually. That probably goes back to the interaction design minor I took in college.) In the end, I put the entry point at the moment a user starts logging a problem, with a short hint suggesting the extension makes this easier.

After the new version shipped, extension installs and problems logged through the extension went up, as I had hoped. With so few users, the change in ratios doesn't prove much. But I could see that the users who had been entering problems by hand had all switched to the extension, which tells me the change worked. I then ran the same loop (observe user behavior, form a hypothesis, make a change, check whether the hypothesis held) to ship Study Plans. They tell you which problems to solve so that each one carries over to many others. Fewer problems, more you can solve. Some users have already started using them.

What this experiment taught me

This was the first time I built a product from scratch entirely on my own and got positive feedback from real users. At the company, my work was largely incremental improvement on a product that already existed. Corporate VPs and above set the big direction, PMs defined the smaller ones, designers and frontend engineers owned UX changes, and dedicated teams handled the backend and infra. I was mostly responsible for the engineering and technical side of the ML solutions. I had no chance to take part in decisions outside engineering: how to acquire users, how to grow, how to sell the product, how to change how the product is presented so users stick around. Splitting roles by function may be efficient inside a company. But what has always motivated me and felt most rewarding is learning across different fields to solve one problem, and then seeing a decision that combines all of it actually win users over. In this CodingPartner experiment, I got to experience all of that.

In the extension story, the same friction showed up twice. The first time, the feature didn't exist yet, and a friend pointed out that manual entry was a hassle. The second time, the feature existed, but users hadn't found it and were still entering problems by hand. That taught me two things:

  • There can still be a real gap between building a feature and getting users to actually use it. The same behavior can have different causes, and each cause points to a completely different fix.
  • An entry point placed at the exact moment users need it works better than one placed where it's most visible.

What I enjoyed most about this one-person project: the whole process was full of uncertainty, and much of the time it felt like finding my way through fog. I could keep trying until the fog cleared, find what the problem really was, and then design a solution for it.

The experiment continues

CodingPartner is still in its cold-start stage. It doesn't have many users, and I don't know yet where it will go. But I want to keep this experiment running and grow along with it.

To me, this blog is a testing ground for life experiments. I want to be a lifelong learner who stays hungry to learn and passionate about life. When a problem bothers me, I treat it as an experiment: break the problem down, propose a solution, test it on a small scale, then put it in front of more people. This post records the first one.

If you're preparing for coding interviews too, I'd love for you to try CodingPartner and tell me what you think. I'd really appreciate your feedback.