ML Systems Lab

You can finish every ML course and still freeze when a model degrades silently in production. Nobody teaches you to debug a stale feature store, or explain a confidence calibration failure, or reason about a latency spike under a new traffic pattern. ML Systems Lab is where you train that. Start with a broken system. Reason through it. See how a senior engineer reads it.

Can you debug it in production? Used by 500+ engineers in interview prep and production triage.

Live → ml-systems-lab-v9xe.vercel.app · Private beta — email for access

---

What makes it different

Pyodide — Python runs in your browser. No server, no setup. Spark jobs, model training, drift calculations execute for real.

Web Speech API — Verbal practice with live transcription. Say your answer out loud. Hear it back. Close the gap between knowing and saying.

StaffLayer — IC3 → IC5 → Staff reveals on the same scenario. The only place that shows you what "Staff-level thinking" actually means in ML.

Bug Hunt — Real production ML code with exactly one buried flaw. Find it before the interviewer does.

---

Flagship experience: Interview zone

Ten simulation tools built around one goal — walk into the room ready.

• Combinator — 45-min timed mock. 100 questions locked until time ends. Full per-domain debrief.
• Defense Plan — Paste a JD → gap map → day-by-day study plan. PDF export.
• Verbal Practice — Web Speech API voice recording · 25 questions · 4-criteria self-rating.
• Interview Q&A — 128 curated MLE questions with model answers and 4-tier scoring.
• Take-Home Bank — 15 open-ended system design questions · model answer reveal · self-score /20.
• Spot the Flaw — 12 adversarial analyses with a buried methodological error. Find it before the interviewer does.
• Trainer — MCQ drill with weakness heatmap and spaced repetition queue.
• Staff Layer — IC3 → IC5 → Staff answer reveals on the same question.
• Bug Hunt — 20 Python/SQL production bugs. One flaw per snippet.

---

What's inside

300+ scenarios · 6 domains · 10 interview tools

ML Engineering

Math Foundations (Pyodide sandbox) · Feature Engineering · Model Evaluation · System Design · Classical ML · Project Lab (Pyodide end-to-end notebook)

Data Engineering

Spark Lab · Airflow · dbt · Data Modeling

Deep Learning

Training Lab · Fine-tuning (LoRA/freeze) · DL Serving (quantization/GPU)

Data Science

Causal Inference · Time Series

MLOps

Monitoring · Deployment · CI/CD & Infra

Read

∇ Gradient — 50 long-form production ML posts linked directly to practice modules

---

Quick start

Live: ml-systems-lab-v9xe.vercel.app Private beta. Email for access.

No account. No install. All progress in localStorage.

# Run locally
git clone https://github.com/SidharthKriplani/ml-systems-lab
cd ml-systems-lab
npm install
npm run dev      # → http://localhost:5173

---

Stack

React 18 + Vite SPA · CSS variables design system · Pyodide (Python in-browser) · Web Speech API · localStorage only · Vercel auto-deploy on push to main

---

Ecosystem

---

Built by Sidharth Kriplani.