Skip to content

Take-home projects and eval take-homes

  • AI coding round
  • Asked often
  • About 1 h 50 min to learn
  • 3 vetted resources
Build a plan that includes this

What this topic covers

Roughly a third of disclosed Applied AI hiring processes include an asynchronous take-home: a 2–7 day window to build a small RAG, agent, or document-extraction system in 2–4 hours, followed by a 45–90 minute defence call. Reviewers are not grading cleverness; they grade scope discipline, a README that records decisions and trade-offs, a baseline plus a small verified test set, an eval harness built before the main logic, basic tests on edge cases, and production awareness (cost, latency, PII, failure handling). Eval-flavoured take-homes add traps: noisy or inconsistent data, composite metrics that hide problems, and open-ended prompts that reward judgment over metric-chasing. The defence round is often weighted more than the code, so every choice must be defensible out loud.

Builds on

You’re ready when you can

What interviewers check, from the resources below.

  • State the typical deadline, expected hours, and what follows the submission
  • List five things reviewers score beyond “does it run”
  • Explain why an eval harness should exist before the main logic
  • Name the three mistakes the guide says sink most submissions
  • Describe how you would budget a 4-hour take-home from reading the brief to submitting
  • List the components of a strong submission (baseline, test set, compared approaches, cost and quality numbers, error analysis, reproducible setup)
  • Explain what a README must cover beyond setup instructions
  • State the rules for using a coding agent on a take-home so you can still defend every choice
  • Give two examples of what an AI take-home brief looks like today
  • Spot two traps a take-home dataset can hide and say how you would surface them in your write-up

What to study, in order

Each link opened and checked by hand; the date says when.

  1. Start here

    Home Assignments (opens in a new tab)

    AI Engineering Field Guide (GitHub) · Alexey Grigorev · 30 min · Free · checked 21 Oct 2026

    Why
    It is the only write-up built from real AI-engineer take-homes (100+ candidate repos, 51 company processes), so it tells you the actual deadline, hour budget, rubric weights, and red flags reviewers use.
    What to read
    Read “Format”, then the “Evaluation Criteria Found in Assignments” subsection, then “How to Prepare”. Skim the H3 archetypes under “Assignment Examples” just to see which shapes (RAG Q&A, agents, extraction) recur.
    What to skip
    The company-by-company lists and the whole “June 2026 Additions” section; they are a catalogue, not advice.

  2. Free alternative

    How to Get an AI Engineer Job in 2026 (opens in a new tab)

    Louis Bouchard (What's AI) · Louis-François Bouchard · 15 min · Free · checked 21 Oct 2026

    Why
    Frames the modern take-home as “ship, measure, defend” and gives a concrete checklist of what a strong submission contains, including how to use coding agents without losing the defence call.
    What to read
    Read “The new interview: ship, measure, defend” (its three numbered subsections) and “What modern AI take-homes look like”. Optionally “Use coding agents like a senior engineer”.
    What to skip
    “The AI engineer portfolio that gets callbacks”, “Choose the AI engineering path you actually want”, “What to build this weekend”, and the FAQ.

  3. Go deeper

    Cracking the AI Evals Interview (opens in a new tab)

    Maven (free lesson) · Hamel Husain · 60 min · Free · checked 21 Oct 2026

    Why
    Evals-flavoured take-homes are a weed-out stage at many AI teams, and this lesson walks through the specific traps (composite scores, noisy labels, no baseline) that separate a pass from a reject.
    What to read
    Read “What you'll learn” and “Why this topic matters” on the page, then watch the full recording using the “In this video” chapter list to pause after the sections on structuring the take-home and on presenting trade-offs.
    What to skip
    “Go deeper with a course” and “Keep exploring”; they advertise the paid course.

Companies that test AI coding round most

Put Take-home projects and eval take-homes in your plan

Tell us your interview date and hours a day; it lands on the right day, in the right order.

Build my plan