Vibe Code a Crypto App
Vibe coding interviews are showing up at more companies.
You receive a prompt, a time limit, and access to AI tools. You need to scope a product and produce something tangible.
Oleh's crypto background and knowledge of Lovable and ChatGPT helped a little bit.
But what made his interview stand out was how he used his limited time. The first half of the interview was for brainstorming and scoping the app. The second half was for vibe coding.
Approaching the problem
Most candidates make the mistake of jumping straight into code generation.
Oleh laid out his stack before writing a single prompt:
"I'm going to be utilizing Notion. I just want us to kind of brainstorm and validate the ideas together so you can see what I'm doing. I'm going to be using ChatGPT. I think it's going to be extremely important for us to get any kind of risks or insights to narrow down the question. And then I'm going to use Lovable. I'm going to try to vibe code and prototype at the end. Once we refine the MVP, once we know that's what we actually want to build."
In a vibe coding interview, your tools should follow your thinking. Declare your stack up front, explain how each tool fits into your workflow, and resist the urge to start building before you've scoped the problem.
Researching with AI
Oleh's first move was to compare stock ownership and crypto ownership in the US. He prompted ChatGPT and found that stock ownership sits around 62% of US adults while crypto ownership hovers around 40%.
That gap became the foundation of his MVP. There's a large population of people comfortable with investing who haven't crossed over to crypto yet.
He asked about regulatory tailwinds, and surfaced the GENIUS Act as a signal that institutional adoption is accelerating and retail will follow. He also asked about risks (regulatory fragmentation, unclear token classification, oversight complexity) and documented those too.
Each query was a building block for the MVP: the opportunity (ownership gap), the timing (regulatory momentum), and the constraints (market fragmentation). In under five minutes, Oleh had a data-backed argument for why this product should exist.
Use AI to build your case, not just your product. Market sizing, trend validation, and risk identification can all happen in real time during the interview.
Making assumptions
Oleh had framed the gap between stock ownership and crypto ownership as a straightforward opportunity. The interviewer spotted the hidden assumption:
"There's an assumption that people that own stocks would potentially want to own crypto."
They both agreed this would need validation in a real product process, but for the interview, they'd move forward with it as a stated assumption.
The leap from "people own stocks" to "those people want crypto" is not self-evident. An interviewer who wanted to be adversarial could have spent ten minutes pulling at that thread.
State your assumptions before someone has to find them for you. When an interviewer surfaces an assumption you missed, acknowledge it cleanly, note what you'd do to validate it with more time, and keep moving. Don't get defensive.
Picking a user segment
With the market framed, Oleh moved to user segmentation. He identified four groups within retail investors: crypto builders, active traders, meme coiners, and Web2 newbies who have never touched crypto.
He chose Web2 newbies because of its large total addressable market, high severity of pain, and the clearest path to his business objective of growing crypto ownership.
This is where the interviewer later flagged a strategic vulnerability. By listing all four segments and only explaining why he chose one, Oleh left the other three as open questions. In the interviewer's words:
"You left the door open on meme coiners, active traders, and builders by not showing why you invalidated them. What you could have done is just said, 'For user segment, I'm going to pick Web2 newbies because they are new to this environment' and left all of the other people out. Then I would have had no questions whatsoever. But because you mentioned all these other people, now I have five questions. I could have completely derailed the whole interview on that."
In an interview, every option you mention is a door you open. An adversarial interviewer can walk through any of them. If you're going to show breadth, you need to also show why you're narrowing.
Addressing pain points
Oleh identified three pain points for Web2 newbies:
- fear of interacting with crypto (volatility, unknown territory),
- unknown concepts that don't resemble anything in their experience (staking, smart contracts, proof of work),
- and regulatory risk (what if it all gets banned?).
He ranked them by severity and frequency, then moved to solutions:
- an AI-assisted agent that helps with crypto interactions,
- an education-driven wallet that explains every step in simple terms,
- and AI-driven crypto news personalized for retail investors.
"For me, everything starts in crypto when you actually hold a token. Now you have unlimited capabilities — to stake, to transfer it, to do anything that blockchain allows you to do. I picked the AI-assisted agent because it can streamline this conversion to buy your first crypto."
If you can get someone past their first purchase, they become a holder. Once they're a holder, education, news, and engagement all have somewhere to land.
The interviewer reinforced this by pushing on leading indicators: "What's an even earlier leading indicator towards somebody who would have a buy conversion?"
Pick the solution that unlocks the most downstream value. Then define the funnel so it's clear exactly where your product creates leverage.
Prototyping
With the problem scoped and the solution chosen, Oleh finally opened Lovable.
Rather than prompting Lovable directly, he used a ChatGPT plugin specifically trained on Lovable's prompting patterns. He pasted his Notion notes into ChatGPT and asked it to generate a detailed user journey, and then fed that output to Lovable.
"There's a nice trick that I really like to use, which is going into ChatGPT, having the Lovable plugin, and then defining the user journey really fast with it. It saves you tens of minutes of trying to define a step-by-step user journey. It just spits it out."
Within minutes, he had a working crypto app with an AI copilot interface. He tested it live and asked the AI to compare Bitcoin and Solana, asking what to buy first, asking where to invest $1,000.
Then he went further. He found a crypto app design screenshot online, uploaded it to Lovable, and asked it to restyle the prototype to look like Robinhood. In about 30 seconds, the entire UI was transformed into something familiar and polished.
"If I ask customers what fintech or brokerage apps they use, and then they see an exact mimic of that for crypto, that gives them confidence. That's something they're really familiar with."
Don't build from scratch when you can build from reference. Using existing design patterns that your target users already trust is a product decision. ChatGPT to generate the prompt and Lovable to generate the prototype is faster than trying to do everything in one place.
Handling user fears
The interviewer raised one of the sharpest challenges of the interview. Oleh was building a product that stacked three things people are afraid of: crypto, AI, and entering credit card information online.
"You have a system where people are afraid of crypto. You're integrating AI, which people have a fear of. And you're also integrating using a credit card online. So now you stack three fears. How do you mitigate that?"
Oleh answered in layers:
- For AI fear: Personalize the agent, let it ask questions at the start so the interaction feels more like a conversation than a black box.
- For payment fear: Partner with a regulated custodian, support familiar payment methods like ACH and Plaid, and make it feel no different from depositing into a stock brokerage.
- For crypto fear: Put guardrails on what the AI can recommend, make the user the one who performs the action ("AI gave me the context, but I'm the one clicking buy"), and connect the experience to concepts they already understand from stock investing.
The user stays in control. The AI educates and recommends. The human decides and acts. That separation builds trust.
When your product combines multiple unfamiliar elements, address each fear individually. And always keep the user in the driver's seat for the highest-stakes action.
Interviewing vs. the real world
After the interview, the interviewer gave the most valuable piece of feedback.
"Because we rushed through the beginning part, I pushed you in a way that actually opened you up to get rejected in an interview because of the doors that we left open."
In the normal course of product work, showing multiple options and then narrowing is exactly the right approach. But in an interview, especially at companies that are filtering hundreds of candidates, every option you mention without invalidating is an invitation for the interviewer to question your judgment.
"I've been in interviews where the interviewer just didn't like the person, and they purposefully went after open doors. They'd say, 'This person didn't think about this.' But the candidate was just focused on where they were going."
In an interview, only open the doors you want to walk through. If you mention alternatives, explain why you're rejecting them. Otherwise, just state your choice with conviction and move on.
Oleh's own reflection aligned with this.
"You have to catch the intent the interviewer has. If you catch the intent that they're allowing you to go and fix yourself, that's one thing. If they allow you to proceed and move forward with your pace, that's another. Try to have a brainstorm session with them. It's not only about you saying 'I'm the best, I'm picking this and that.' You give objective arguments, but at the same time you're asking, 'What do you actually think?'"
Treat every interview as a collaboration, but be strategic about what you expose. In a vibe coding round especially, you're moving fast and making lots of choices. State your choices with conviction, close the doors behind you, and save the breadth-of-thinking showcase for when you can fully defend it.
What makes a great candidate?
What made Oleh stand out had nothing to do with the quality of the prototype:
"He jumped in and was himself. He showed off his skills. He showed his workflow. There was a point where I jumped ahead and he was like, 'Yeah, I have this, it's right here.' He handled it so smoothly. I felt the slap on my wrist, but it was so nicely handled. It was very vibey and collaborative."
Oleh didn't panic or abandon his structure. He redirected the interviewer back to his outline politely.
This is Oleh's personal framework for vibe coding interviews, which he uses to self-evaluate:
-
Originality: What unique insight or mechanism am I creating? The AI-assisted first-buy experience wasn't technically novel, but the insight that first purchase unlocks everything else gave it strategic depth.
-
Impact: What's the quantified impact? Oleh tied everything back to his business objective (double crypto ownership) and defined the conversion funnel that would measure progress.
-
Cross-domain bridging How am I connecting tech, business, and design? The Robinhood-style UI wasn't a cosmetic choice. It was a behavioral science decision to reduce friction.
-
Influence — How does this work influence stakeholders and customers? The prototype itself becomes a communication tool. It's higher definition than a PRD and is faster to iterate than a full build.
Interview tips
Before the interview
Decide on your tool stack and know how the tools connect. Oleh's Notion → ChatGPT → Lovable pipeline was rehearsed, not improvised. Practice the full workflow end-to-end at least once before your interview.
During the scoping phase
Use AI as a live research partner. Document everything in a shared workspace so the interviewer can follow your thinking. Make assumptions explicit and move forward. Don't wait for permission to decide.
When narrowing options
Only open doors you intend to walk through. If you list alternatives, explain why you're eliminating them. If you can't defend the elimination, just state your pick and commit. Every unexplained option is an invitation for the interviewer to challenge you.
During the build phase
Layer your tools. Use ChatGPT or Claude to generate detailed prompts, then feed them to your code generator. Use reference designs from products your users already trust. Don't try to make the prototype perfect.
When the interviewer pushes back
Read the intent. Are they trying to help you close gaps, or are they stress-testing your conviction? Either way, acknowledge the challenge, add nuance, and keep driving. The worst thing you can do is freeze or abandon your direction at the first sign of resistance.
The interviewer isn't evaluating whether you can vibe code a perfect app. They're evaluating whether you can think clearly under pressure, make deliberate product bets, and use AI tools to accelerate your judgment.