I built an AI coding mentor that never forgets your mistakes — using Hindsight Memory.

Keshav Anand — Fri, 20 Mar 2026 18:08:41 +0000

Have you ever practiced coding on a platform, closed the tab,
and came back the next day to the exact same random problems?
No memory of what you struggled with. No personalization. Just
starting from zero again.

That's the problem I wanted to solve at the Hindsight Hackathon.

What I Built

CodeCoach AI — a personalized coding mentor that remembers
every mistake you make, detects your weak topics, and generates
problems specifically targeting your gaps — across every session.

How Hindsight Memory Powers It

Hindsight by Vectorize is a memory system for AI agents. It has
three core operations that I used throughout CodeCoach:

1. retain() — After every attempt

Every time a student submits code, the result gets stored:

The AI evaluates the code, identifies the specific mistake,
and stores it as a memory. Not just "they got it wrong" — but
"they forgot the base case in a recursion problem" or "they used
wrong syntax in a function definition."

2. reflect() — Before generating a problem

Before creating a new problem, CodeCoach asks Hindsight to
reason over the student's full history:

Hindsight synthesizes all past memories and returns something
like: "This student struggles with recursion base cases and
Python function syntax." The AI then generates a problem that
specifically targets those weak areas.

3. recall() — For the Learning Profile

The "My Learning Profile" button shows everything Hindsight
remembers about the student — every mistake, every topic,
every pattern detected over time.

This is the demo money shot. Watching the memory graph fill up
in Hindsight Cloud as a student practices is genuinely exciting.

The Architecture

The app has 3 Python files:

memory.py — all Hindsight communication (retain, recall, reflect)
mentor.py — all Groq AI communication (generate problems, hints, evaluate)
app.py — Flask web server that connects everything

Every user gets their own persistent memory in the Hindsight
bank. The more they practice, the smarter their mentor gets.

Tech Stack

Hindsight Cloud — persistent memory system
Groq (qwen/qwen3-32b) — fast AI for problem generation
Python + Flask — backend
HTML/CSS/JS — frontend

What Makes It Different

Without Hindsight, this would just be another random problem
generator. With Hindsight, it becomes a mentor that genuinely
knows you — your patterns, your struggles, your growth.

After 3 practice sessions, CodeCoach knows:

You struggle with recursion base cases
You make syntax errors in function definitions
You're ready to move from easy to medium difficulty

No other coding platform does this.

Try It

GitHub: https://github.com/keshavanandtezz-bit/Codecoach-Ai
Live Demo: https://codecoach-ai-7yl2.onrender.com

Built at Hindsight Hackathon 2026

This was my first hackathon. If you're thinking about joining
one — just do it. You'll learn more in 24 hours than in weeks
of tutorials.

hindsight #python #ai #hackathon #webdev

DEV Community: Keshav Anand