Introduction
If you're anything like me, you've spent hours trying to use AI for interview preparation.
A few months ago, I started using Claude seriously. Honestly? It was impressive — much better than ChatGPT or Gemini for breaking down complex system design concepts and algorithms. But as my preparation stretched over weeks, the cracks started showing.
The context window filled up and responses became painfully slow. There were no visual aids to show me what I had mastered versus what I was still failing at. And worst of all zero persistent memory. Every single new session, I had to re-explain myself from scratch:
"I'm applying to Google. My weak area is System Design. Here's the 30-day study plan we made last week. Here are the companies I'm targeting..."
It was exhausting.
Standard LLMs have a hangover problem. They wake up every morning remembering absolutely nothing from the night before.
That frustration is exactly why I built Interview Brain.
What is Interview Brain?
Interview Brain is an AI-powered interview preparation platform I built for the WeMakeDevs x Cognee Hackathon 2026.
The core idea is simple but powerful: instead of relying on a stateless chat window that forgets you every session, Interview Brain uses Cognee's hybrid graph-vector memory to build a living, persistent knowledge graph of your entire career profile.
Your skills. Your weak areas. Your target companies. Your mock interview history. Your progress over time.
All connected. All remembered. All yours.
The Problem with Standard AI Tools
Before I explain what I built, let me paint the picture more clearly.
Here's what using Claude for interview prep actually looked like after a few weeks:
Session 1: "I'm targeting Google and Sarvam AI.
Weak at System Design."
→ Great response ✅
Session 2: "Remember my study plan?"
→ "I don't have access to previous conversations." ❌
Session 3: Context window full.
→ Responses getting slower and less accurate ❌
Session 4: "What companies am I targeting again?"
→ Has to explain everything from scratch ❌
The AI wasn't getting smarter about me. It was staying exactly as dumb as the day we met.
The Solution: Cognee's Knowledge Graph
Cognee solves this by moving your context out of the LLM prompt and into a persistent hybrid graph-vector memory layer.
Think of it like giving your AI a proper brain — one that builds actual relationships between concepts, not just stores chunks of text.
Here's how I used all 4 core Cognee memory lifecycle operations in Interview Brain:
- remember() — Never Explain Yourself Again When you sign up on Interview Brain, you fill in your profile:
Name and target role
Skills you have
Companies you're targeting
Areas where you're weak
Interview Brain immediately calls Cognee's remember() to ingest all of this and build your initial knowledge graph.
Khushhal → has skill → FastAPI
Khushhal → has skill → RAG
Khushhal → weak at → System Design
Google → asks about → System Design
Google → asks about → Distributed Systems
Khushhal → targeting → Google
From this point on — you never repeat yourself. The graph remembers everything.
recall() — Personalized Everything
This is where Interview Brain gets genuinely impressive.
Prep Roadmap: You select a company and number of days. Interview Brain uses Cognee's recall() to traverse your knowledge graph and generate a day-by-day study plan that specifically targets YOUR weak areas against THAT company's known interview style.
Not generic advice. Not "study arrays and linked lists." Actual personalized guidance like:
"Day 3: Focus on CAP theorem and distributed consistency — you've struggled here before, and Google specifically tests this in system design rounds."
Mock Interview: Interview Brain generates questions by recalling your profile + target company data from the graph. It speaks questions out loud using voice synthesis, records your answer via voice recognition, then evaluates your response against what Cognee knows about that company's expectations.
The result? A score, detailed feedback, and your graph automatically updates with the new performance data.visualize() — See Your Brain
One of my biggest frustrations with standard AI was the complete lack of visual feedback.
Am I actually improving? What skills do I have? How do they connect to my target companies?
I integrated Cognee's native /visualize endpoint with a D3.js force-directed graph to create the Skill Map — a live, interactive visualization of your entire knowledge graph.
You can literally see:
How your skills connect to target companies
Which topics are your strongest nodes
Where the gaps are in your preparation
I also added an Enhance button that natively triggers Cognee's improve() operation to enrich and strengthen the graph connections as you learn more.
- forget() — Stay Clean Dropped a company from your target list? No problem. One click on Interview Brain calls Cognee's forget() to surgically remove that company's dataset from your memory graph. No stale data. No irrelevant recommendations. Your graph stays accurate and up to date.
Technical Architecture
For those who want to go deeper, here's how everything fits together:
Frontend (Vanilla JS)
↓
FastAPI Backend
↓ ↓
SQLite DB Cognee Cloud API
(users, (knowledge graph,
feedback, vector search,
sessions) graph traversal)
Why this separation?
SQLite handles structured queries — "give me all feedback where outcome = weak." Cognee handles intelligent queries — "given this candidate's profile and Google's requirements, what should they focus on?"
Both together = a real product, not just a fancy chatbot.
Security:
JWT authentication on every protected route
bcrypt password hashing
Per-user isolated Cognee datasets (interview_brain_user_{id})
No cross-user data leakage possible
The Mock Interview — The Wow Moment
If I had to pick one feature that makes Interview Brain stand out, it's the mock interview.
Here's what happens:
- Select company (e.g. Google)
- Camera turns on
- AI speaks question out loud
- You answer by voice
- AI evaluates your answer against:
- Company's known interview style (from graph)
- Your specific weak areas (from graph)
- Best practices for that topic
- Score + detailed feedback
- Graph automatically updates
- Next question It feels like sitting in an actual interview — not chatting with a bot.
What I Learned
Building Interview Brain taught me something fundamental about AI product design:
The bottleneck isn't the LLM. It's the memory.
Claude, GPT-4, Gemini — they're all incredibly capable. But without persistent, connected memory, they're like hiring a genius consultant who gets amnesia every morning.
Cognee flips this. By separating memory from inference, you get:
Speed (graph traversal is fast)
Personalization (your graph, your data)
Visual understanding (you can see the connections)
Evolution (the graph grows with you)
This is what EdTech AI should look like. Not a chatbot. A coach that actually knows you.
Try It Yourself
If you're preparing for interviews and tired of explaining yourself to AI every single session — give Interview Brain a try.
GitHub: https://github.com/Kh9705/interview_brain
Demo: https://interview-brain.onrender.com
And if you're a developer — the architecture is clean and the Cognee integration is well documented in the README. Fork it, adapt it, make it your own.
Built solo for WeMakeDevs x Cognee Hackathon 2026
Stack: FastAPI + SQLite + Cognee Cloud + Vanilla JS
By Khushhal Sinha
Final Year CSE, LNMIIT Jaipur
Top comments (0)