“I think you’ve made this mistake before.”
That line caught me off guard.
I didn’t tell the AI anything about my past attempts.
There was no chat history. No manual input.
And yet, it recognized a pattern.
That’s when I realized — this wasn’t just another coding assistant anymore.
The Problem I Kept Running Into
While solving problems, I noticed something frustrating:
I wasn’t stuck because problems were hard.
I was stuck because I kept repeating the same mistakes.
- Writing incorrect loop conditions
- Missing edge cases
- Using inefficient approaches And every time?
The AI responded like it was the first time seeing me.
Same mistake → same generic feedback.
No memory. No improvement.
Rethinking What a “Mentor” Should Be
Most AI tools follow this pattern:
You ask → AI answers → conversation ends
But that’s not how learning works.
A real mentor:
- remembers your weaknesses
- notices repeated mistakes
- adapts explanations over time So I asked:
What if my coding assistant could remember how I code?
Step 2: Recalling Past Patterns
Before generating feedback, I fetch relevant memories:
The system analyzes past behavior and generates context-aware insights.
Now the AI doesn’t just respond to code…
It responds to the user behind the code
What Actually Changed
Here’s the real difference.
Before Memory
“There is an issue in your loop.”
Correct — but generic.After Memory
“You’ve struggled with loop conditions before — check if this is an off-by-one error.”
Same code.
Same model.
But now the feedback feels aware.What I Didn’t Expect
I thought memory alone would solve everything.
It didn’t.
Problems I faced:
- too much memory = noisy responses
- raw logs = no meaningful insights
So I improved it by:
- tagging mistakes (syntax, logic, performance)
- limiting recall results
- summarizing memory before sending to AI
How the System Behaves Now
After a few attempts, the system starts to:
- recognize repeated mistakes
- avoid repeating feedback
- adjust explanations
- guide instead of just correcting
It doesn’t “know” the user perfectly.
But it knows enough to feel different.
What This Project Taught Me
- Memory is more important than model size
- Personalization comes from patterns, not prompts
- Less but relevant context > more context
- AI becomes useful when it becomes consistent
What’s Still Missing
There’s still room to improve:
- smarter pattern detection
- prioritizing important memories
- long-term skill tracking
Try it
https://codemento.netlify.app/
Final Thought
Most AI tools today feel smart.
But they forget everything.
This project made one thing clear:
The real upgrade isn’t making AI smarter.
It’s making it remember.
Built at HackWithIndia 2026- Aastha Sawade
Team: Krishna Hasare, Karan Patil, Priya Vhatkar, Siddhi Shinde,
Sarvesh Gajakosh
Top comments (0)