๐ ๐ฆ๐ต๐ถ๐ฝ๐ฝ๐ฒ๐ฑ ๐ฎ๐ป ๐๐ ๐ฆ๐ธ๐ถ๐น๐น ๐ฎ๐ ๐ฎ ๐ฃ๐ฎ๐ถ๐ฑ ๐ช๐ฒ๐ฏ๐๐ถ๐๐ฒ
I turned an internal skill into a real product.
It is not a demo or a prototype. It is a live website where users sign in, pay, and get results. The product is Hidden Stems. It is a tool for Bazi, or Chinese astrology. Users enter birth details and get a chart with a deep reading.
Building this required more than just a web form. I had to set up:
- User accounts and sign-in flows
- Stripe checkout and webhooks
- Database persistence
- Vertex AI for generation
- Vercel deployment and environment variables
- SEO and basic configuration
- Error handling for payments and AI
In the past, the last 10% of a project felt heavy. I would reach 90% completion and then stop. The remaining work felt boring and full of traps. Deployment, configuration, and production errors used to kill my momentum.
This year, the process felt different.
I used Codex to help manage the technical stack. I used the following tools:
- Codex for reading code, editing, and fixing bugs
- Stripe CLI for payment verification
- gcloud CLI for Google Cloud and Vertex AI
- Vercel CLI for deployment
- Supabase for the database
The cloud setup used to feel intimidating. I felt I needed to understand every concept before touching it. Now, it feels like debugging a path forward. Codex walks with the CLI. It reads errors and suggests the next command.
Stripe integration also became easier. Instead of guessing why a payment failed, I could check the flow in small steps:
- Is the provider configured?
- Do the live keys exist?
- Does the webhook endpoint work?
- Did the database update?
The engineering loop is much faster now. But the human work has not disappeared. It just moved.
The hardest part was not the code. It was the frontend.
A working page is not the same as a page that makes people pay. I spent a lot of time on:
- Making Bazi understandable for English speakers
- Balancing Chinese terminology with global design
- Making the paid report feel valuable
- Ensuring the mobile layout works perfectly
AI can help you finish the engineering. It can suggest UI directions. But it cannot replace product judgment. You still need to guide the visual taste and the level of trust.
Shipping no longer feels like one giant wall. It feels like a pile of small problems. If you have tools to help you break those problems down, your projects will not die at 90%.
How do you handle the final polish of a web product? Do you design first or ask the AI for directions? I would love to hear your workflow.
Optional learning community: https://t.me/GyaanSetuAi