100+ AI take‑home repos

A GitHub collection of more than 100 real AI take-home repositories is giving candidates and hiring managers a rare look at what companies actually ask people to build. (github.com) The collection sits inside Alexey Grigorev’s “AI Engineering Field Guide,” a public repository he describes as research into AI engineering roles, interviews, and hiring practices from Q4 2025 and Q1 2026. The repo’s README says its interview section includes take-home assignments and paid work trials drawn from 100-plus GitHub repos. (github.com 1) (github.com 2) Grigorev’s March 9, 2026 webinar notes say the take-home analysis examined recent submissions from late 2025 and early 2026 to map current hiring standards. An earlier March 3, 2026 webinar previewed the same material as patterns from “100+ GitHub repos,” alongside job descriptions and candidate reports. (github.com 1) (github.com 2) A take-home assignment is a build-at-home test: a company gives a candidate a small product problem, the candidate ships code and notes, and then defends the choices in a follow-up interview. Grigorev’s guide says these assignments are usually asynchronous, come with 2-to-7-day deadlines, and are commonly followed by a 45-to-90-minute walkthrough. (github.com) The guide ties those repos to a wider hiring dataset. It says the home-assignment section is based on 1,765 job descriptions, 100-plus GitHub repos of candidate submissions, and practitioner reports; among 51 companies with disclosed interview processes, 17 included a take-home or asynchronous assignment and 5 more used paid work trials. (github.com) The assignments cluster around a few repeated tasks. Grigorev’s analysis says more than 40% of the repos involved retrieval-augmented generation, or RAG, which is a way to make a chatbot fetch documents first and answer from those files instead of from memory alone. (github.com) Agentic systems accounted for 30% or more of the repos, conversational AI for 20% or more, document-processing tasks for about 15%, and “LLM-as-judge” evaluation setups for more than 10%. The examples in the guide include PDF question-answering bots, policy assistants with mandatory citations, and systems that must say they do not know when the retrieved context is missing an answer. (github.com) The pattern in those prompts matches the rest of Grigorev’s interview research. His hiring guide says interviewers increasingly test evaluation frameworks, cost and latency trade-offs, observability, safety, and tool use with coding assistants, rather than only model training theory. (github.com) The take-homes also show how companies are handling AI tools unevenly. Grigorev’s guide says only one company in the sample explicitly allowed AI tools in take-home assignments, none explicitly banned AI for take-homes, and most did not state a policy at all. (github.com) For candidates, the practical value is simple: the market is publishing its homework. In one public repo, the prompts, code samples, time windows, and defense-round expectations are now visible in the same place. (github.com)