Vibe Coding in Practice: How a Finance Graduate Built a $1K/Month Interview Assistant with Zero Code and AI

What Is Vibe Coding? A New Programming Paradigm

Vibe Coding refers to a way of building software entirely through natural-language collaboration with AI: you don't write code, you just describe intent, and AI (such as Claude or Cursor) generates and iterates on the implementation. This article is based on a real livestream by a YouTube creator, showing how he built an AI interview assistant app from scratch using this approach.

The creator was originally a finance graduate who taught himself programming for only six to seven months, and now independently develops websites, mobile apps, and desktop software. His core tool stack is: ChatGPT for planning, Claude for coding, Google Stitch for designing, and the OpenAI API and Firebase for backend capabilities.

During the livestream, he openly admitted he was still learning OBS streaming, and even couldn't clearly show his screen operations because his microphone was broken. This "learning by doing" state is exactly a true reflection of how low the barrier to Vibe Coding is—you don't need to become an engineer first before you can start building products.

The Project: An "Invisible" AI Interview Assistant

Product Positioning and Core Features

The product he is polishing is called Hidax, an AI interview assistance tool. Its key selling point is being "invisible to screen sharing and screen recording software," making it suitable for remote job interviews, online defenses, and similar scenarios. Its features include:

• CV/resume personalization: After uploading your resume, AI generates answers based on your experience.
• Research project defense support: After uploading papers or project materials, AI learns them and gives personalized answers to reviewers' questions.
• Pay-per-use: A $5 interview package lasting about 90 minutes, aimed at users who only need a single high-stakes interview.

Invisibility to screen sharing is typically not achieved through a single technique, but relies on OS-level window capture exclusion mechanisms. For example, on macOS, an app can call the exclusion window API provided by ScreenCaptureKit so that its own window is not captured by screen recording or sharing tools; on Windows, SetWindowDisplayAffinity can mark a window as "capture-prohibited." Such mechanisms are system-level APIs rather than the app "hiding" its window itself, so they generally work against software like Zoom, Google Meet, and OBS that follows system capture interfaces. It should be noted that this cannot counter physical camera filming or standalone capture cards; the "invisibility" in the product's marketing should be understood as invisibility to mainstream screen recording and sharing software, not absolute undetectability.

Pricing Strategy and Early Data

The product has been live for about a month and currently has 45 registered users, of whom about 15 have subscribed to the $30/month Premium plan while the rest are still on trial. It also offers a $5 one-time payment option, targeting people who "don't need a full-month subscription and only need to handle one interview." According to the creator, this is a tiered pricing design for users with different payment capacities.

Breaking Down the Development Process: Security Gates from Planning to Deployment

Toolchain and Division of Labor

• ChatGPT: Requirement breakdown, feature planning, and release roadmap organization.
• Claude / Cursor: Actual code generation and debugging.
• Google Stitch: Interface and visual design.
• Firebase: Backend, database, and cloud functions.
• Lemon Squeezy: Payments and subscription billing.

Firestore security rules are the access control layer of the Firebase database, using declarative rules to define "who can read and write which data under what conditions." Unlike traditional backend code, the rules themselves are evaluated on every database request, so getting the rules wrong means the frontend can bypass business logic and tamper with data directly. The "security fixes and Firestore rule consistency checks" mentioned in the original are essentially about confirming that the permission model assumed by the client code exactly matches the rules actually enforced by the database, avoiding vulnerabilities such as users being able to upgrade their own subscriptions or read other people's data.

Atomic transactions refer to a group of database operations that either all succeed or all roll back, with intermediate states invisible to other requests. In the subscription billing scenario, "deducting usage quota" and "activating subscription entitlements" must be placed in the same transaction; otherwise, if the second step fails, the user ends up in an inconsistent state of "money paid, permissions not enabled." Firestore provides the runTransaction interface to support this semantics, but requires developers to proactively wrap the relevant reads and writes inside it—which is exactly why the original emphasizes "using atomic transactions to synchronize usage and subscription entitlements."

The "Security Gate" Before Release

One detail he repeatedly emphasized during the livestream was the security validation before launch. He was doing local verification for the new admin billing system to ensure administrators can manually upgrade subscriptions for users—preventing failures where "the user paid but the plan wasn't activated." Specific checks include:

1. Local build and testing of the new admin billing UI;
2. End-to-end validation of Premium and pay-per-use features;
3. Security fixes and Firestore rule consistency checks;
4. Ensuring support admins cannot arbitrarily modify subscription entitlements or activate legacy plans;
5. Using atomic transactions to synchronize usage and subscription entitlements, avoiding inconsistency.

He mentioned that there are currently 45 passing emulator tests in the isolated release environment and that he approved the cloud function deployment, but still postponed the formal deployment because the previous production revision was not fully captured.

This caution is not excessive: for an independent developer, a billing error directly equals lost revenue. Vibe Coding lowers the barrier to writing code, but it does not lower the requirements for system design, permission control, and data consistency.

The Real Barriers and Opportunities of Vibe Coding

Where the Barriers Are

This creator's path shows that Vibe Coding is not "zero-barrier": he spent six to seven months studying almost every waking hour before reaching the level where he could independently deliver websites and apps. The real difficulty is not syntax, but:

• Breaking vague requirements into executable steps;
• Judging whether AI-generated solutions are secure and maintainable;
• Understanding systemic issues such as deployment, permissions, and billing.

Lessons for People Without a Technical Background

He has now delivered about 10 websites and several mobile apps, with clients including small businesses. His advice is straightforward: don't shelve product ideas just because you "can't write code"; you can first use AI to learn how to build, or find someone who understands AI development to collaborate with. The real value of AI tools is not helping you write emails, but turning ideas into runnable, monetizable products.

Conclusion

This late-night livestream had no polished demo—just a self-taught developer debugging a billing system, fixing security rules, and dealing with a microphone failure. The Vibe Coding picture it presents is pragmatic: AI handles a large amount of coding work, but product thinking, system security, and continuous iteration are still the responsibility of humans. For ordinary people who want to get into the game, this may be the most worthwhile path to reference.

One-line Summary

A finance career-changer used Vibe Coding to build an AI interview assistant in one month, earning about 15 paying users per month.

Paragraph Summary

The article documents how a creator with a finance background, who taught himself programming for only six to seven months, built the AI interview assistant Hidax from scratch through Vibe Coding (driving AI to generate code with natural language). The toolchain is clearly divided: ChatGPT for planning, Claude/Cursor for coding, Google Stitch for design, Firebase for backend, and Lemon Squeezy for payments. The product's selling point is being "invisible to screen sharing"; after about a month online, it has gained 45 registered users, with about 15 subscribing to the $30/month Premium plan and a separate $5 pay-per-use option. The article focuses on breaking down the pre-release security gates, including billing UI testing, Firestore rule consistency, permission control, and atomic transaction synchronization, and points out that Vibe Coding lowers the coding barrier, not the capability requirements for requirement breakdown, system security, and deployment judgment.