100+ AI take-home repo list

A public GitHub guide has turned more than 100 real AI take-home assignments into a searchable study set for candidates and hiring managers. (github.com) The collection sits inside Alexey Grigorev’s “AI Engineering Field Guide,” a repository that says it is based on Q4 2025 to Q1 2026 interview research, 100+ GitHub repos of candidate submissions, and 1,765 job descriptions for the home-assignment section. (github.com) In that sample, 17 of 51 companies with disclosed interview processes used a take-home or asynchronous assignment, and five more used paid work trials. The same guide says take-homes are usually completed in two to seven days and followed by a 45- to 90-minute defense round. (github.com) An AI take-home is a timed build-at-home test: a company gives a candidate a small product problem, and the candidate ships code, a write-up, or a demo. In these repos, the most common version was retrieval-augmented generation, or RAG, which means a chatbot first searches documents and then answers from the retrieved text. (github.com) The guide’s breakdown says 40%+ of the sampled repos centered on RAG systems, 30%+ on agentic systems that call tools in multiple steps, 20%+ on conversational assistants, 15% on document processing, and 10%+ on “LLM-as-judge” evaluation. One example prompt asks for a document question-answering system with citations and a refusal when the answer is not in the source material. (github.com) Those patterns line up with the same repository’s wider job-market analysis. Its March 2026 update says the project had compiled 2,445 AI engineer job descriptions from January through March 2026, and a separate role summary says RAG appeared in 35.9% of jobs while agents appeared in 14.4%. (github.com 1) (github.com 2) The assignment list also makes one hiring-policy gap visible. The home-assignment guide says only one company in its sample explicitly allowed artificial intelligence tools in take-homes, no company explicitly banned them there, and most did not state a policy at all. (github.com) The repo is not a neutral census of every AI interview. Grigorev describes it as a research project built from public repos, practitioner reports, blogs, discussion threads, and scraped job postings, which means it reflects what candidates and companies exposed in public rather than private hiring flows. (github.com 1) (github.com 2) Even with that limitation, the list gives candidates something rare: a way to inspect the actual shape of recent AI hiring homework instead of guessing from generic interview advice. It shows that, in early 2026, many companies were testing less for model training and more for shipping document search, tool use, evaluation, and production-style tradeoffs. (github.com 1) (github.com 2)