AI Engineer Interview

Stop reading about the answer. Get scored on it.

Every guide here ends in a free puzzle that grades your answer as a percentage of a provably optimal solution — context packing, RAG tuning, routing, orchestration, caching. No login to see your score.

◆ Live · with interactive
Context Window Optimization: 2026 Interview Guide

The context-window interview is a 0/1 knapsack. Learn the relevance-per-token model that beats greedy's 76% — free, no-login puzzle included.

Read + play →
◆ Live · with interactive
Agent Orchestration: 2026 System Design Interview

Designing a production agent is min-cost planning, not fewest tool-calls. Learn the model — free, no-login puzzle that scores your plan vs the optimum.

Read + play →
◆ Live · with interactive
RAG Chunking Strategy: Choosing Chunk Size (2026)

“Just use bigger chunks” fails — chunk size has a sweet spot under a latency budget. Learn how to choose it, with a free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
LLM Caching: Prompt Cache vs KV Cache (2026)

Prompt (prefix) caching vs KV caching, and the ordering rule that cuts your token bill. Free no-login puzzle scores your prompt layout vs the optimum.

Read + play →
◆ Live · with interactive
Tool Routing: Match the Model to the Task (2026)

Don’t route every call to the biggest model — match capability to need. Learn cheapest-capable routing, with a free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
Agent Memory Compaction: Keep, Summarise, Drop (2026)

Under a context cap, summarising a bulky-useful turn beats keeping it whole or dropping it. Learn the model — free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
Diagnosing RAG Failures: Retrieval vs Generation (2026)

Is your RAG answer wrong because retrieval missed the chunk, or because the model ignored it? Diagnose first, then fix — with a free no-login retrieval puzzle.

Read + play →
◆ Live · with interactive
LLM-as-a-Judge: Building an Eval Set (2026)

Evals are the new system design. Pick a representative eval set for marginal coverage, not raw size — with a free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
The Multi-Agent Handoff Interview: When to Hand Off, When to Hold

Hand off only when the specialist’s saving beats the handoff cost — never-switching and always-switching both overpay. Free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
Guardrail Latency Budgets: Which Checks Are Worth Running?

Guardrail catch rates compound, so a few decent checks clear a high safety bar — running everything is wasted latency. Free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
Retry vs Fallback: The Expected-Cost Math of Flaky LLM Calls

A cheap flaky call backed by a reliable fallback beats one premium call — in expected cost. Learn the ordering math, with a free no-login puzzle scored vs the optimum.

Read + play →
◆ Live · with interactive
The Dispatch Problem: Assigning N Agents to N Tasks

Assigning N agents to N tasks for minimum total cost is a matching problem — per-agent greedy collides on shared favourites. Free no-login puzzle scored vs the optimum.

Read + play →
Or jump into all 41 puzzles →