I kept running into the same problem: every time I wanted to try a different LLM provider, I had to rewrite large parts of my code. Different SDKs, different APIs, different response formats. It slowed down experiments and made simple comparisons painful.

So I ended up building a small abstraction layer that lets me switch between providers without touching the rest of the app. One interface, multiple backends.

In this post, I explain:

the exact problem I kept hitting

what design choices worked and what didn’t

how I structured the abstraction

and the trade-offs of this approach

If you’ve ever wanted to compare models or avoid vendor lock-in, this might be useful:

https://dev.to/chief_yaml_officer/how-i-stopped-rewriting-my...

Source: https://github.com/Nayjest/lm-proxy