I'm an independent developer from Ethiopia, and one of the projects I've been building is Lowwin, a mobile-first Telegram Mini App for auctions.
This project has been a practical way for me to explore how AI-assisted development can support real product building. I already have experience with Mini App development, so my workflow is not just “ask AI to make an app.” Instead, I combine my own product and engineering knowledge with ChatGPT Pro and Codex to move from idea to implementation, then testing and iteration.
In this post, I want to share how I’ve been using Codex while building Lowwin, what has been useful, and what I’ve learned from the process.
Note: Lowwin is still an active development project. The screenshots and examples here are development previews.
What is Lowwin?
Lowwin is a Telegram Mini App auction platform designed with a mobile-first experience.
The goal is to create a smooth interface where users can:
- Browse auction items
- View auction details
- Place bids
- Navigate a simple mobile-friendly experience inside Telegram
Some of the current UI areas include:
- A homepage with featured auction items
- Auction listing screens
- Item detail pages
- User-focused navigation for bids, alerts, favorites, deposit, and profile
My workflow: ChatGPT Pro + Codex + my own development knowledge
The most useful workflow for me has been a combination of:
- Thinking through the feature or problem
- Using ChatGPT Pro to improve and structure prompts
- Using Codex to implement, refine, and test
- Reviewing the output as a developer
- Repeating until the result is good
This is important because AI works best when the developer gives it:
- Clear context
- Clear goals
- Expected behavior
- Constraints
- A way to verify the output
I don’t treat Codex like a magic button. I treat it like a very powerful development assistant.
What Codex helped me with
Codex helped me across the Lowwin project from start to end in different ways.
That includes:
- Translating feature ideas into implementation steps
- Generating and refining interface code
- Improving structure and consistency
- Helping with testing-related thinking
- Speeding up repetitive development work
- Making it easier to move from concept to working UI
A big part of my process was using ChatGPT Pro to prepare better prompts, then using those prompts to guide Codex more effectively.
That made a big difference.
Instead of writing vague requests, I could be more specific about:
- What screen I wanted
- How the behavior should work
- What should be improved
- What needed to be tested
- What constraints to preserve
Example: building a real product, not just a demo
One thing I value in AI-assisted development is that it should help build real software, not just attractive demos.
While working on Lowwin, I’ve been focused on building something that feels like a real product:
- Mobile-friendly
- Clear navigation
- Consistent layout
- Usable screens
- A product flow that makes sense inside a Telegram Mini App
That means checking more than just “does the code run?”
It also means asking:
- Does this screen make sense for users?
- Is navigation clear?
- Does the interface feel consistent?
- Does the feature behave the way I intended?
- Can I build on top of this later?
Codex helps a lot, but I still need to make the product decisions.
Screenshot preview
Lowwin homepage
This screen presents the main brand, featured products, and core navigation.
Auction detail screen
This view shows the product, auction information, and bidding entry point.
Auction listing screen
This screen helps users browse available auctions and quickly take action.
What I learned
1. AI is most useful when you already understand what you're building
Because I already have experience with Mini App development, I can guide the process better.
That makes AI more effective.
2. Good prompting matters
A better prompt gives better output. For me, ChatGPT Pro has been useful for shaping prompts before sending them into Codex.
3. Codex speeds up development, but review is still necessary
Codex can accelerate implementation, but I still review structure, logic, UX, and output quality.
4. Building is iterative
The real value is not a one-shot response. It’s the cycle of:
idea → prompt → implementation → review → test → improve
Why this matters to me
As a developer from Ethiopia, I’m especially interested in showing that AI tools like Codex can help builders create real products, not just experiments.
I want to share more practical examples from my own work, especially for:
- Ethiopian developers
- Independent builders
- People building Telegram Mini Apps
- Developers learning how to use AI tools productively
For me, the exciting part is not just using AI — it’s using AI to build things that are actually useful.
Final thoughts
Lowwin has been a good example of how I like to work with AI.
I bring the direction, product understanding, and development experience.
ChatGPT Pro helps me think more clearly and improve prompts.
Codex helps me build faster, refine ideas, and move toward working software.
That combination has been powerful for me, and I’m excited to keep sharing what I learn.
If you're also building with Codex, I’d love to hear how you use it in your workflow.
AI disclosure: I used AI tools to help refine my workflow, organize ideas, and support the writing and development process described in this article.
