I made my AI agents self-improve with real human feedback : dead-simple - no retraining, no hacks.

LLM agents today don’t learn from corrections. You tell them “use percentages, not counts” — next time, they forget.

So I built something to fix that:

Self-Learning Agents → [github.com/omdivyatej/Self-Learning-Agents]

A simple wrapper that makes agents improve automatically after feedback.

Just 2 lines:

pip install dead-simple-self-learning

learner.save_feedback("Summarize contract", "Include indemnity clauses if mentioned.") 
enhanced_prompt = learner.apply_feedback("Summarize contract", base_prompt) 

No retraining. No finetuning. No vector DB.

How it works:

• Embeds each task (OpenAI/MiniLM) -

• Finds similar past tasks

• Retrieves mapped feedback

• Optionally filters with LLM - Enhances system prompt before running

• Works with OpenAI, DeepSeek, LangChain agents.

  If you're building agents and want them to improve with real-world feedback, try it. Would love feedback, ideas, or brutal honest!