Your A/B Test Said +3%. Here's Why the Launch Flopped.
Statistical significance is not a shipping decision. On p-values, practical significance, and the guardrails most teams skip.
Read the essay →Deliberate practice for tech careers
Know what you actually know.
Answer real interview questions, get scored against a rubric, and watch your personal concept graph show you exactly what's solid — and what's still fuzzy. 306 questions across product, data science, data engineering, and AI.
Free plan: 50 AI sessions · No card required · Pro €9/month
getprepgraph.com/arenaExample session
ARENA · DATA SCIENCE · MACHINE LEARNING
Why is accuracy a bad metric for fraud detection?
Type or speak your answer. Get scored against the rubric — not a generic AI take.
“With 99.9% legitimate transactions, a model that predicts ‘not fraud’ every time is 99.9% accurate — and catches nothing. I’d use precision, recall and PR-AUC…”
Needs WorkOn TrackStrong ✓Outstanding
Your concept graph42 concepts
● Mastered● Understood○ Surface
306
questions in the Arena, browsable without an account
4 tracks
Product Management, Data Science, Data Engineering, AI Product & ML
29 topics
from Kafka and dbt to metrics, experimentation and RAG
Every.
answer scored against a rubric — never a vague “good job”
Try it — no sign-upDon't read about it.
Don't read about it.
Answer one.
This is the loop. One real Arena question, scored the way every question in PrepGraph is scored.
PRODUCT MANAGEMENT · METRICS & ANALYTICS
Your A/B test shows a 3% lift with p = 0.04. Do you ship it?
Walk through your reasoning. Mention sample size, practical significance, and what could go wrong. Two or three sentences is enough to try it.
Full feedback, hints and the model answer unlock on the free plan.
Your rubric feedback will appear here.
Needs Work On Track Strong Outstanding
Needs Work On Track Strong Outstanding
How it worksReading isn't knowing.
Reading isn't knowing.
Practice is.
“Interview readiness is a side effect. Knowing is the goal.”
01 — ARENA
Practise in the Arena
Pick a concept from your track. Answer in writing or by voice, under real interview conditions.
Type answerVoice answerHints, one at a time
02 — RUBRIC
Get scored against a rubric
Every answer is graded on what a strong answer actually contains — what you covered, what you missed, and why it matters.
Needs WorkOn TrackStrongOutstanding
03 — WORKSHOP
Track mastery in the Workshop
Concepts are earned, not self-reported. Spaced repetition resurfaces each one right before you'd forget it.
Surface→ Understood→ Mastered
The suiteNot a question bank.
Not a question bank.
A practice system.
Five modules that compound: every gap you find becomes a concept, and every concept you master makes the next answer — and the next match — sharper.
Arena — 306 questions, 4 tracks
Real interview questions with hints revealed one at a time, rubric scoring, and a model answer after every attempt. Fully browsable before you sign up.
84Product Management
93Data Science
72Data Engineering
57AI Product & ML
Workshop
Your personal knowledge base. Add concepts from the Arena, add them manually, or paste an article, paper or transcript and extract 3–8 concepts automatically.
Mastered · 18Understood · 14Surface · 10
Example figures, not live data
Graph
See how your concepts connect — the clusters you've built, the bridges you're missing, and the gaps an interviewer will find before you do.
ExperimentationML SystemsStreamingGap: System Design
Match a Role
Paste a job description and your resume. You get a match score, your strengths mapped to the role, and the critical gaps — each one addable to your Workshop in a click.
78 / 100 · Senior PM, Fintech
3 critical gaps found · 2 already in your review queue
Example result, not live data
MCP × Claude
Your knowledge graph, inside Claude.
Connect PrepGraph to Claude once. Claude can then read your concepts, mastery levels and practice history — and tell you what to study next, in context.
- 1Open Claude → Settings → Integrations
- 2Add the PrepGraph integration URL below
- 3Ask Claude what to study next. Revoke anytime.
Integration URL
getprepgraph.com/api/mcp
getprepgraph.com/api/mcp
Privacy, plainly: Claude can read your Workshop concepts, mastery and practice history. It cannot see your notes or payment info.
CLAUDE · EXAMPLE EXCHANGE, PREPGRAPH CONNECTED
YouBased on my PrepGraph graph, what should I study this week?
Claude · via PrepGraph MCPYou have Kafka and dbt covered — both Mastered. Your Spark is thin (Surface, last practised 19 days ago) and it's blocking two Data Engineering concepts behind it.
Here's the next two weeks: Spark fundamentals first, then Streaming System Design. I've added both to your Workshop review queue.
Here's the next two weeks: Spark fundamentals first, then Streaming System Design. I've added both to your Workshop review queue.
YouAdd RAG evaluation as well — it's in the JD I'm targeting.
Claude · via PrepGraph MCPDone. RAG Evaluation added to your Workshop, linked to your LLM Systems cluster. Your Match a Role score for that JD would move from 78 → 84 once it's Understood.
The Playbook
The ideas the questions test, written up.
Essays on the concepts behind the corpus — the depth that makes the questions worth answering.
Your ML Model Passed the Test. Here's Why It Failed in Production.
Offline metrics, leaky features, and the quiet ways a model that aced evaluation falls apart on live data.
Read the essay →Your Model Isn't the Moat. Your Knowledge Is.
Models commoditise. A personal graph of what you've actually mastered — portable, private, compounding — doesn't.
Read the essay →PrepGraph launches MCP
Your Workshop, inside Claude: what the integration can do, what it deliberately cannot see, and how to connect it.
Read the launch →Pricing
Start free. Stay because it works.
No card to start. No course to finish. Just practice that compounds.
Free
€0
For trying the loop properly
- 50 AI practice sessions
- Intro + intermediate questions in every track
- Rubric feedback on every answer
- Workshop + concept graph
Pro · Most complete
€9 / month
For serious, sustained preparation
- Unlimited AI practice sessions
- Advanced questions in every track
- Match a Role — unlimited checks
- MCP × Claude integration
Built by the founder
Marjuk Ahmad
Integrity Systems at Meta, protecting WhatsApp from automated attacks. Previously marketplace analytics at Uber. He built PrepGraph full-stack, solo, from Dublin — while working full-time — because no prep tool closed the gap between reading something and actually knowing it.
“You’ve read the articles. The question is whether you can answer, under pressure.”DublinFull-stack, soloMeta / WhatsAppEx-Uber
Questions
Before you start
How much does PrepGraph cost? +
Free gives you 50 AI sessions and intro + intermediate questions in every track, with no card required. Pro is €9/month for advanced questions and unlimited sessions.
Which roles and tracks does it cover? +
Four tracks — Product Management (84), Data Science (93), Data Engineering (72), AI Product & ML (57) — across 29 topics, 306 questions in total.
How does the scoring actually work? +
Each question has a rubric: the points a strong answer must make. Your answer is graded Needs Work, On Track, Strong or Outstanding against that rubric, with hints and a model answer — and the result updates your mastery level for that concept.
What can Claude see through the MCP integration? +
Your Workshop concepts, mastery levels and practice history — enough to plan what you study next. It cannot see your notes or payment info, and you can revoke access anytime in Claude Settings → Integrations.
Is this just interview prep? +
No. Interviews are the forcing function. The Workshop and spaced-repetition queue are built for staying sharp in the role long after you've landed it.
Stay relevant. Keep learning. Track your mastery.
Start with one question. See exactly what you know — and what you only thought you knew.
Start practising free →Free plan: 50 AI sessions · No card required