What should a take-home assignment for developers ask for?
A take-home assignment for developers should ask for one piece of the role's real work, sized to finish in one sitting, with each file you want back named. Naming the files tells the candidate what done looks like, and gives you one piece of evidence per skill the role needs.
It comes after the first-round screen: screening a Java developer covers the questions and the short code task that come before it.
Start from the client's job description. Pick the two or three skills a first screen can't show, such as designing a deployment, writing a pipeline or planning a rollout, and set a task that needs exactly those. Leave out everything else.
Then name each file and what it is evidence for. For a developer who also runs the platform, that might be:
- the Kubernetes manifests or Helm values, for containers;
- an outline of the infrastructure code, for the cloud setup;
- the CI/CD pipeline, for how code reaches production;
- a README with the rollout plan, a rough monthly cost and the trade-offs, for judgement.
Ask for the reasoning in writing. The README is the file a recruiter who doesn't write code can read, and compare across candidates.
Keep the task to a design or a small slice. A task that looks like a ticket from the client's backlog reads as unpaid work, and the better candidates turn it down.
Take home assignment for BDE
A take-home for a business development executive is a written task: a short plan for a named territory or account list, and two or three outreach emails.
A strong answer names its targets and gives a reason for each one. It sets out a sequence with the timing between steps, keeps the emails short enough to read on a phone, and commits to a number the candidate would hold themselves to, such as meetings booked or replies. The rule from the developer task holds: write down what good looks like before you send it. A written plan is a document like any other file, so it goes back the same way.
For the first-round questions on a close sales role, see the SDR screening questions.
How do you score a take-home assignment?
Score a take-home against a list of what a strong answer contains, written before it goes out, and score each skill from the files that show it. A single overall mark hides which part was strong, so tie each file to a skill on the role's scorecard. Who goes to the client is still your call.
Write the list as items you can check. For a deployment task it might read: health checks and resource limits on the service, secrets kept out of the repository, a pipeline that builds, tests, scans and can roll back, a rollout in stages, a cost with its working, and the trade-offs named.
Read the README first. It shows whether the candidate made decisions or listed tools. A string of product names with "zero downtime" beside them says less than one choice with its reason and what it costs.
Then check the claims against the files. A candidate who writes that they run large clusters, and whose manifest has no health checks, has handed you two things that don't agree. Ask about the gap before you decide.
For the code you can't read, check three things: every file you asked for is there, the pipeline has a test step and a way back, and the README explains the choices. Leave the engineering judgement to whoever reads the code, a person or a tool.
Put the items into a weighted interview scorecard. Each skill gets a weight and a 1 to 5 scale with words for each level, and the take-home becomes evidence for some of those skills, next to the interview answers.
What should the take-home brief tell the candidate?
The brief should tell the candidate how long the task should take, when it is due, each file to send and in what format, and what will be read. Candidates fit a take-home around a job, so a clear brief gets you more finished answers and fewer that guess at the scope.
Put these in the brief:
- Time and deadline, as two things. Say how many hours the task should take, and give a window of days to send it in.
- The files, by name and purpose, and the formats you accept. Ask for the files themselves, not a ZIP, so whoever reads them can open each one. Code can go into Markdown or text files.
- What will be read. Name the skills the take-home is evidence for. Keep the scoring words to yourself.
- AI tools. Say whether they are allowed, and that the candidate may be asked to explain a choice in a follow-up.
- Payment and use. Say whether the work is paid, and that it won't go into the client's product. Agree both with the client first.
- A contact. Name who answers questions, and say when the candidate will hear back.
When is a take-home the wrong step?
Skip the take-home when you approached the candidate yourself, when the client runs a live technical round anyway, or before the candidate has passed the client's hard conditions. A take-home asks hours of a candidate, so it belongs after the knockout questions and in place of a round, not on top of one.
- Headhunted senior candidates and engineering managers. Ask about their own systems instead, and leave the build to the client's round.
- A client whose engineers run a pair-programming or system-design round next. The take-home would repeat it.
- Before the knockout questions. Notice period, location and the client's other hard conditions take a minute to check. Don't ask anyone for hours of work who fails one of them.
- The client's real backlog. That is unpaid work, and good candidates decline it.
Where it sits is the recruiter's and the client's call. It can be a separate step after the first screen. It can also be one question inside the same interview, when the candidate answers in their own time and can stop and come back.
How Recagn does this
Recagn is our product. We ran a senior Java role through it: Recagn drafted the rubric and seven questions from the job description, and we added a code question and one take-home. The four candidates are invented, and their answers and take-home files were written for this example. Recagn scored them on 17 Nov 2026.
The take-home as a Document question
In Recagn, a take-home is a question answered as a Document: the candidate uploads files instead of typing.
- A candidate can upload up to 10 files per answer, each up to 10 MB and 20 pages: PDF, DOCX, images, Markdown, text or ZIP. An old .doc file is refused with "Save it as DOCX or PDF".
- Recagn reads the text of every file. Markdown, text and DOCX files are read directly, and PDFs and images go through OCR. Files are read in the order they were uploaded, up to 40 pages for one answer. A ZIP is listed by name and not opened, which is why the brief above asks for the files themselves.
- The question is linked to the rubric criteria it is evidence for, and Recagn scores each criterion from every answer linked to it, uploaded files included. There is no separate take-home score.
- Candidates have a deadline window of 3 to 30 days from the invite, 14 by default. In a self-paced interview they can stop and come back within it.
Here is more on coding screening for recruitment agencies in Recagn.
The brief and what came back
The take-home we set read:
"Take-home: you are moving one Java Spring Boot service from a VM to Kubernetes on AWS (EKS). Upload your design as up to five files: the Kubernetes manifests or Helm values, a Terraform outline for the cluster and its database, the GitLab CI/CD pipeline, and a README with your rollout plan, rough monthly cost and trade-offs."
Before scoring, we wrote down what a strong answer contains: "A plan that could be deployed: health probes, resource requests and limits, secrets kept out of git, the cluster, database and network in Terraform modules, a pipeline with build, test, scan and deploy stages and a rollback, a staged rollout, a realistic cost estimate and named trade-offs."
The question is evidence for three criteria on the scorecard, which Recagn calls a rubric: AWS & infrastructure as code (20%), Containers & platform engineering (20%), and Linux administration & CI/CD (15%).
Four answers to the same take-home
| Candidate | Files uploaded | The cost line, in their words |
|---|---|---|
| Rohan D. | 5: README.md, k8s.md, terraform.md, gitlab-ci.md, cost.txt | "About USD 520 a month against USD 610 for the two VMs plus their standby, mainly from right-sizing and scaling down at night." |
| Sneha K. | 3: README.md, helm-values.md, pipeline.md | "I don't have a good number." |
| Vikram S. | 2: Architecture.md, infra.md | "Expected to be significantly lower than the current setup." |
| Arjun M. | 1: answer.md | "I am not sure about the cost. I would check the AWS pricing page." |
Source: Recagn worked example (Senior Java full stack and platform engineer), scored 17 Nov 2026 · candidates invented; files written for this example
The cost line is the one item a recruiter can compare without reading any code. Across four answers to one brief, it runs from an estimate with its working to "not sure".
The answer ladder: Containers & platform engineering
These are the rubric's words for each score on Containers & platform engineering, beside a line from each candidate's take-home and Recagn's reasoning. The score is for the criterion and rests on more than the take-home.
| Score | The rubric's words | Candidate | From their files | Recagn's reasoning, in part |
|---|---|---|---|---|
| 5 | "Platform engineering expert; runs enterprise clusters and has built/maintained Backstage plugins and internal developer portals." | Rohan D. | "Rollback at any step is moving the ALB weight back to the VMs, which takes under a minute and needs no deploy." (README.md) | "The EKS take-home adds mature production-grade manifests (probes, PDB, topology spread, HPA, IRSA)." |
| 4 | "Operates production K8s/OpenShift; strong with Docker/Podman and some Backstage/IDP work." | Vikram S. | "Blue-green deployment with zero downtime. Traffic is shifted gradually using Istio. If anything goes wrong we roll back." (Architecture.md) | "The take-home shows awareness of Karpenter, Istio, ArgoCD and GitOps. However, the detail is thin and claim-heavy … and the EKS manifest uses image:latest with no probes or resource limits." |
| 2 | "Uses Docker; limited Kubernetes exposure and no platform-engineering work." | Sneha K. | "I have deployed services to Kubernetes with Helm, but I have not built an EKS cluster or written its Terraform." (README.md) | "The honest, well-structured take-home demonstrates competent application-level K8s knowledge but not cluster operations, keeping it at 2." |
| 1 | "No container orchestration experience beyond running a local container." | Arjun M. | "I have not used Kubernetes, EKS or Terraform before, so I could not write the manifests or the Terraform properly." (answer.md) | Doesn't mention the take-home: "No container orchestration experience and Docker only heard of, not used in real work." |
Source: Recagn worked example (Senior Java full stack and platform engineer), scored 17 Nov 2026 · candidates invented; files written for this example
Each score rests on every answer linked to the criterion: Containers & platform engineering also draws on the cluster-administration question and the production-incident question.
Rohan D.'s rollback line names the mechanism and how long it takes. Vikram S.'s says "we roll back". Vikram S. claims large platforms, and Recagn's reasoning sets the manifest he uploaded against those claims, which kept him at 4. That is the README check from earlier, on real files.
Where they landed
Recagn recommends Qualified for Rohan D. and Maybe for Vikram S., against the rubric's lines of 3.8 for Qualified and 3 for Maybe. Their weighted totals, 4.40 and 3.10, are worked out from the criterion scores and weights by the rule in how AI interview scoring works.
Sneha K. and Arjun M. get no total here, because the hard gates decide both (a knockout question is what Recagn calls a hard gate). Recagn's reason for Sneha K. ends "failing three of four gates", and for Arjun M. it says he "fails all technical gates". Recagn recommends Disqualified for both.
Both did the take-home: Sneha K. uploaded 3 files and Arjun M. 1. They were still disqualified on the experience and skill gates, judged in the same interview. That is why the knockout questions belong before the take-home.
Reading every uploaded file and scoring each criterion with its reasons come off your plate. What the take-home asks for, where it sits in the process, and who goes to the client stay yours. Recagn recommends. Only you can reject a candidate.
Questions people ask
Can a candidate get ChatGPT to write a take-home on Recagn?
They can, and the take-home is read for it. Recagn reads each answer, uploaded files included, for signs it was written by AI. Activity tracking (tab switches, copy and paste) also records when files are picked for upload, on any interview whose questions were locked after activity tracking launched, which covers every interview you set up now. Both sit beside the score, never in it, and you decide whether they matter. A follow-up asking the candidate to explain one choice in their README tests it further.
How is a Recagn take-home different from a HackerEarth coding test?
HackerEarth's help centre says "The generated score is the sum of the scores of the test cases that the candidate’s solution has passed" (HackerEarth, checked 17 Nov 2026). Recagn reads a take-home's files, code (not run), README and cost estimate against the role's rubric, so a design task with no single right output can still be scored.
What does an interview with a take-home cost on Recagn?
The take-home adds nothing: uploaded files add no credits, however many there are, so with no video answer and webcam proctoring off the whole interview is one credit, at most ₹50 on Starter before GST (₹49.98; ₹58.98 with 18% GST). Asking any question for a video answer raises it to three credits, at most ₹150. Both figures are upper limits, since the same credits pay for reading CVs.
Should a take-home assignment be paid?
Pay for it, or keep it small, when it would take more than one sitting or looks like the client's real work. Agree it with the client before sending, and say it in the brief, along with the fact that the work won't be used.