Practical Labs

Real bugs in real-world data. You are handed a wrong number and a live dataset — investigate with SQL, find the root cause, and prove the fix. Everything runs in your browser.

75 labs ready across 12 lab tracks

🧱Pro

Data Engineer

Pipelines, warehousing, and SQL correctness under real-world dirty data.

“Revenue is double-counting on a customer join”“Late-arriving facts break an incremental load”“Timezone drift misplaces rows across daily partitions”
6 labs ready · 6 total
📊Pro

Analytics Engineer / BI Developer

Metric models, dashboards, and the subtle ways numbers go wrong.

“A metric definition change makes MoM growth lie”“Fan-out inflates revenue in a one-to-many dashboard join”“NULLs collapse a segment out of the report”
6 labs ready · 6 total
⚙️Pro

Backend / SQL-heavy Engineer

Transactions, query performance, and data-integrity bugs.

“A lost update left an account balance wrong”“A count forgets the soft-delete flag”“Inclusive BETWEEN double-counts the month boundary”
6 labs ready · 6 total
🤖Pro

ML / AI Engineer

Feature pipelines, data leakage, and evaluation traps.

“Target leakage from a feature recorded after the label”“The train/test split leaks the same user into both”“Accuracy looks great because the labels are imbalanced”
5 labs ready · 5 total
🔬Pro

Data Scientist / Analyst

Analysis pitfalls: aggregation paradoxes, sampling, and bias.

“Simpson's paradox flips a conclusion after aggregation”“Survivorship bias hides the churned population”“Repeated testing manufactures a fake significant result”
5 labs ready · 5 total
🕹️Pro

Agentic AI Engineer

Debugging LLM agents in production: tool-call integrity, loops, cost, and memory.

“Retried tool calls double the side effects”“Orphaned tool calls break the next model turn”“A broken stop condition inflates step counts”
10 labs ready · 10 total
✨Pro

Generative AI Engineer

RAG grounding, eval sets, latency and token spend — the numbers that flatter a GenAI system.

“Three sources turn out to be one document sliced three ways”“Similarity scores compared across two embedding models”“The hallucination rate excludes every request that went worst”
10 labs ready · 10 total
🚨Pro

Site Reliability Engineer (SRE)

Debugging production: error rates, cost budgets, logs, metrics, and traces.

“The checkout error-rate SLO is reading high”“Monthly cloud spend blew the budget overnight”“The error-count widget over-reports failed requests”
6 labs ready · 6 total
🛠️Pro

DevOps Engineer

Debugging CI/CD pipelines and Kubernetes: flaky runs, OOM kills, pod scaling, and memory.

“CI pass rate is inflated by workflow re-runs”“Pipeline duration is inflated by parallel jobs”“OOMKilled count is overstated by total restarts”
6 labs ready · 6 total
☁️Pro

Cloud Architect

Cost attribution, reservation coverage, and multi-region numbers that do not add up.

“Shared platform cost counted twice in the org total”“Failover makes the global request count look higher than it is”“Reservation coverage report says the org is over-covered”
5 labs ready · 5 total
🏛️Pro

Data Architect

Dimensional modelling under pressure: SCD2, bridges, conformed dimensions, and key stability.

“Inventory on hand grows every day the pipeline runs”“Industry revenue adds up to more than the company earned”“Half of European revenue disappeared when billing was onboarded”
5 labs ready · 5 total
🧭Pro

DevOps Architect

DORA metrics and SLOs that mislead: composite availability, deploy counts, and cluster headroom.

“Deploy frequency is measuring how often the pipeline fails”“Change failure rate counts staging breakages against production deploys”“MTTR stops the clock at the first resolution, not the last”
5 labs ready · 5 total