My Coding Agent Remembered Which Language Trips Me Up — And Quietly Steered New Challenges Away From It.

I kept noticing something frustrating.

No matter how many times I fixed the same bug, AI tools would treat it like the first time.

That’s when I realized the problem:
stateless AI can’t help you grow.

What I Built Instead

I built CodeMentor around one principle:

Every mistake should matter in the future.

Stack:

• React frontend
• Groq for inference
• Hindsight for memory Loop: Recall → Analyze → Retain

The Missing Piece: Memory

Before analysis, I fetch past mistakes:

const mems = await hs.recall(bankId, "coding mistakes");

Then feed them into the model.

Now feedback becomes contextual.

Learning Through Repetition (and Breaking It)

The system:

• Detects repeated mistakes
• Highlights patterns
• Adjusts responses

This turns debugging into learning.

Storing Experience

await hs.retain(bankId, `Repeated issue: ${mistakes}`);

Now mistakes aren’t wasted—they’re reused.

Final Thought

Stateless AI answers questions.

Stateful AI builds skill.