hey everyone 
i’ve been playing around with langchain agents for a while, and honestly got tired of this cycle:
agent fails → check logs → tweak prompt → test → repeat
it felt like i was the one learning, not the model 
so i hacked together something called DeltaLoop
it just takes the logs your agent already produces and turns them into training data, then fine-tunes a small LoRA adapter automatically.
repo’s here if you want to check it out or play with it:
github.com/M4T1SS3/DeltaLoop
also:
i’m still pretty new to this community, so if anyone has feedback, ideas, or wants to contribute, i’d really appreciate it 