Advice on full-stack project for a non-coder?

I’ve been interested in making my own RSS aggregator with some custom features (user authentication, ability to save/share profiles, keyword filtering,
AI API for advanced filtering, email digests).

I attempted this a few months ago with Cursor (purely by prompting) and ended up with a front-end only client that relied on an external RSS parsing API and configurations saved into the browser cache.

Recently I tried again with Replit and got a nicer looking full-stack app, but stalled halfway there (AI becomes hopelessly unable to fix bugs, add new features).

I preferred the experience of using Cursor, but I’m not confident about being able to make and deploy a full-stack app as a non-technical person. What I liked about the first experience was being able to paste the entire code into other LLMs when I hit roadblocks, but this doesn’t seem as doable with a modular architecture.

Any advice on planning for this, more suitable frameworks, where to deploy, etc.?
The Replit project used the following (not that I have any attachment to this stack).

· React with TypeScript for the frontend
· Express.js for the backend
· PostgreSQL for data storage
· TanStack Query for data fetching
· Shadcn/ui for the user interface components

I’d recommend creating a Project at https://claude.ai/ and talking to the model there about what you want to build and all the questions you have about how to architect.

Then from those conversations, you can ask Claude to draft and refine specifications for the project. This will help you have a better idea of how you’ll approach the project and fill in your knowledge gaps. When creating specs, ask Claude to create them as artifacts, so you an see your chat in one column and the spec in another.

From there, you can bring the specs into Cursor and use those as requirement docs for your buildout. I typically create a requirements/ directory and store each spec there as markdown file. This is the approach I take for any medium to large sized project. The requirement docs are invaluable for keeping Cursor on track. And the convo with Claude is a great way to do initial planning no matter your skill level.

4 Likes

It’s gonna be tough, because Cursor (or any AI model directly) is gonna make mistakes here and there. I use Cursor for my dev job and two things happen frequently:

  • AI makes small mistakes that I point out and make it fix, or just quickly fix myself
  • AI makes bigger mistakes or fails to grasp a situation well, so I have to mostly just do it myself

For both of those scenarios I am only able to notice the AI has made a mistake because I have the coding knowledge to see them.

Another thing worth mentioning: I’ve found the AI works best when I clearly outline the work that needs to be done. Understanding the task and explaining it from a high level perspective, then allowing AI to handle the finer implementation details. This sort of “high level architectural knowledge” is what you’ll absolutely have to have, even if you are weak on the finer points of coding syntax and techniques.

Suggestion by @devgrinder is a good one. Talk to a different AI model directly about what you’re trying to do, it can help you brainstorm, build the framework, project roadmap, break things into individual tasks. It can also help you format instructions to be most effective when fed to Cursor.

Luckily coders will still have a job for another year or so… I’ve seen mine revert code, add bugs, straight up not understand, or even worse… say it knows how to do something and it doesn’t at all and wastes an hour of my time…

I’d suggest learning test driven design at bare minimum. Write the tests in plain text, then have it incrementally add to it.

I think that’s the only way to safely do anything no-code style that’s remotely complex.