How to Build a Coding Assessment From Scratch (Without Reinventing Leetcode)
Start from the job, not the puzzle
The first mistake most teams make: they pick a problem that seems hard. The right starting point is a job description and the answer to one question — what does this person need to do in their first 90 days?
If the answer is "fix bugs in our Django service," your assessment should look like fixing bugs in a Django service. Not balancing a binary search tree.
Step 1: Write the job-to-be-done
Three sentences. "By month three, the candidate should be able to: (1) ship a small feature end-to-end with tests, (2) debug a production issue with logs, (3) review a teammate's PR with substantive comments."
This is your assessment design brief.
Step 2: Build the smallest possible repo
Spend an afternoon building a real-but-tiny version of your stack: a Postgres table, a couple of API endpoints, three failing tests, a deliberate bug or two. Total size: 200–500 lines. Open source it if you can — this is great recruiting marketing.
Step 3: Pick a scope, hard
A 90-minute window is the sweet spot. Long enough to do something real, short enough that candidates do not feel their weekend was hijacked. Anything over 4 hours and you will lose strong candidates with families and current jobs.
Step 4: Write the rubric before the prompt
Four dimensions, anchors at each level, no more. (See our structured scorecard template.)
Step 5: Pilot internally
Run two of your own engineers through the assessment. Time them. If a senior engineer takes 70% of the time budget, the budget is right. If they finish in 20 minutes, your problem is too small. If they cannot finish, it is too big.
Step 6: Pair it with a live followup
Always. The take-home is the artifact; the followup is where you confirm the candidate authored it. The followup should be 30–45 minutes, on the candidate's actual submission, with the prompt: "walk me through this and let's add one small thing."
Step 7: Ship it via a platform with integrity signals
Self-hosting on GitHub works for one role. For a hiring funnel, you want a platform that delivers the prompt, captures keystroke and screen telemetry, runs AI-assisted grading, and surfaces the result to the hiring manager. ClarityHire ships exactly this loop with a one-click clone-from-template.