Hey everyone, I’ve been playing around with Replit for a while now and honestly, it’s the BEST no-code app builder out there. Saying this as someone with absolutely zero coding experience — never written a single line of code. Huge props to the Replit team.
I’m posting here because I’m a little confused about the next step. I’m building an academic web app for students that helps humanize AI-written text and fixes grammatical mistakes.
How do I decide which API to go with? Should I use Claude’s Sonnet or ChatGPT’s API? And do I need to train it, or will it do the job on its own?
A bit stuck on this step… any help or resources would mean a lot.
Welcome! Sounds like you have some exciting ideas.
On the topic of choosing which AI API to go with, you can really go with either if you are building with Replit Agent as there are integrations for either Anthropic Claude or OpenAI API from ChatGPT. More info there on that: Replit Docs
As far as deciding which to go with, that is challenging to answer as they both have their strengths for sure - but one thing to consider is test both out and see which one gives better results for your app!
Best wishes and it would be fun to see what you end up with.
Training it is an interesting topic and something that I believe could end up being unique for each app. Typically when I have built apps with AI integrations there will be some kind of a ‘system prompt’ in the app that instructs the app and AI via API how it should respond. Its challenging to break down the process here in the forum, however, if you want to see a public example for reference to see how the integration works, I do have an app that was published in the Replit Gallery recently that uses OpenAI API integration: Journey Mapper - Replit Gallery
I agree with @mark, for most applications, the LLM itself w/out custom training and a finely tuned system prompt does an excellent job. In fact, it usually outperforms custom-trained models as long as you’re doing a general enough task. For your specific example, the vast knowledge of the LLM is potentially a benefit and not a drawback. Until you get into it and see how it performs, you won’t know for sure.
Keep coming back and letting us know how its going!