Introduction
Over 6 months of interviewing, I went from scattered notes and forgotten topics to a system that tracked everything — 100+ ML topics, every company pipeline, and every round in one place. Then I open-sourced it.
The Problem with ML Interview Prep
Every ML engineer preparing for interviews eventually hits the same wall.
You open a browser tab for “ML interview questions.” Then another for “system design ML.” Then another for LeetCode. Then a YouTube playlist on transformers. Then a blog post on RAG. Then someone’s GitHub repo with 200 starred links.
Three hours later, you have 47 tabs open and zero structured knowledge.
The resources exist — they’re scattered across hundreds of blog posts, courses, and repos. But no single system connects them. You can’t see what you know, what you don’t know, and what matters most for your next interview.
I tried:
- Notion databases — became a graveyard of half-finished pages
- Markdown notes — couldn’t see the big picture
- Anki flashcards — great for memorization, terrible for understanding relationships between concepts
- Other people’s repos — always missing the topics I needed, or organized in ways that didn’t match how I think
"The problem isn't a lack of resources. It's the absence of a system that grows with you."
Building the System
What started as messy interview notes evolved into something much larger. Over months of real interviews — some that went well, many that taught me hard lessons — I built a system with five components:
1. Knowledge Base — The Core Reference
A single, massive Markdown file covering 100+ ML/AI/DS/LLM topics across 14 categories. But unlike typical study guides, every topic has a two-layer structure:
- Quick reference tables — one-liners, key formulas, interview phrasing you can actually say out loud
- Detailed explanations — derivations, code examples, edge cases, “what interviewers are really asking”
The topics aren’t just textbook summaries. They’re battle-tested against real interview questions. When an interviewer asked me something I couldn’t answer well, I went back and added or expanded that topic.
The knowledge base is a living document — it grew with every interview, not before them.
2. Interactive Mind Map — See Everything at Once
The knowledge base is for depth. The mind map is for breadth.
I built a D3.js visualization that renders all 100+ topics as an interactive, searchable, zoomable map. You can see at a glance how topics relate — how Transformers connect to Attention, how RAG connects to Vector Databases, how Feature Engineering connects to everything.
Click the image above to explore the live interactive mind map
Features:
- Search — find any topic instantly
- Focus mode — click a topic to see its relationships
- Dark mode — because we all study at night
- Export — save as PNG for quick reference
- Related topics — navigate between connected concepts
The mind map isn’t hand-drawn. It’s auto-generated from metadata in the knowledge base using a sync script. Edit the markdown, run make kb-sync, and the map updates.
"The knowledge base is for reading. The mind map is for seeing. The sync script is the bridge."
3. Interview Tracking System
I built a template system with 10 explicit design decisions — from how to name files to how to handle transcripts to how to separate prep from post-interview review. Five reusable templates cover the full lifecycle:
- Research template — company intel, team context, tech stack
- Round template — prep, live notes, post-interview review
- Prep tracker — rolling checklist across all rounds
- Quick reference — consolidated study material per company
- Postmortem — cross-round analysis after a pipeline completes
The system is designed so that 15 minutes after an interview, you can do a brain dump and have everything you need for an AI-assisted debrief later.
4. Career Frameworks
Beyond technical prep, interviews require strategic thinking:
- Offer decision matrix — structured framework for comparing offers across dimensions (compensation, growth, team, culture)
- Negotiation strategy — tactics and scripts for salary negotiation
- STAR story methodology — templates for behavioral interviews with a systematic approach to crafting stories
- Career strategy — “Domain-First Prep” and “T-Shape Audit” frameworks
5. AI Prompt Templates
Every workflow I used during prep — debriefing after interviews, drilling before them, setting up tracking for a new company — I converted into platform-agnostic prompt templates. They work with Claude, Gemini, ChatGPT, or any LLM.
The Mind Map: Seeing What You Know
The 14 categories span the full ML interview surface area:
| Category | Example Topics |
|---|---|
| ML Foundations | Gradient Descent, Bias-Variance, Cross-Validation |
| Classical Algorithms | Decision Trees, SVM, KNN, Naive Bayes |
| Ensemble Methods | Random Forest, XGBoost, Gradient Boosting |
| Deep Learning | CNN, RNN, Attention, Batch Norm |
| NLP & LLMs | Transformers, Tokenization, Fine-tuning, RLHF |
| Agentic AI | LangChain, LangGraph, RAG, Tool Calling |
| MLOps | CI/CD, K8s, Monitoring, Feature Stores |
Open-Source
I’ve open-sourced the entire system:
ml-interview-prep-kit — Knowledge base + mind map + templates + tooling
What you get:
- 100+ topic knowledge base with interview-ready phrasing
- Interactive mind map at interview.prasanth.io
- 5 interview tracking templates with a worked example
- Career frameworks — offer matrix, negotiation, STAR stories
- AI prompt templates for prep workflows
- Build tooling — mind map sync, link validator, PII scanner
The PII scanner is worth highlighting: when you add your own notes — real company names, real interview details — the scanner catches confidential terms before you accidentally commit them.
# Get started in 3 commands
gh repo fork prasanth-ntu/ml-interview-prep-kit --clone
open knowledge-base/ml-ds-llm-fundamentals.md
make mindmapClosing
Built with Claude Code and D3.js. Open-sourced at ml-interview-prep-kit. Live mind map at interview.prasanth.io.