# ARIS-in-AI-Offer: a bilingual AI interview cheat sheet collection built by an agent workflow > A Python repository of 34 Chinese and English ML, LLM, diffusion and agent interview cheat sheets, generated by the ARIS /interview-cheatsheet and /render-html skills and shipped as single-file HTML, plus a CV-to-academic-homepage generator. **wanshuiyin/ARIS-in-AI-Offer** — Bilingual (中文+EN) ML / LLM / diffusion / agent interview cheat sheets for AI 秋招 — generated by ARIS /interview-cheatsheet, rendered by /render-html into single-file HTML, reads anywhere — plus a CV→DBLP-fact-checked academic homepage generator and hand-authored long-form blogs 🌱 - Repository: https://github.com/wanshuiyin/ARIS-in-AI-Offer - Stars: 547 · Forks: 19 - Language: Python - License: MIT - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/wanshuiyin-aris-in-ai-offer ## The gap ARIS-in-AI-Offer fills in Chinese AI recruiting prep Chinese AI campus recruiting, the season the README calls 秋招, compresses a wide technical surface into a short window. Candidates need derivations, runnable code and a question bank that matches what interviewers ask, and the repository is explicit that it is aimed at that audience. The README describes each cheat sheet as a long-form Chinese tutorial with three pillars: formula derivations, from-scratch PyTorch code, and 25 high-frequency interview questions stratified into L1 essentials, L2 advanced and L3 top-tier lab. The collection spans 34 first-party cheat sheets across 7 categories plus one community-contributed category, covering foundations, post-training and reasoning, LLM architecture and systems, generative model theory and tokenizers, generation systems for image, video, 3D and diffusion post-training, multimodal models, agents, and embodied AI. The second audience is narrower but real. The repository is not only a study aid, it is a demonstration of the ARIS workflow. The README states that every cheat sheet here is the production output of the same /interview-cheatsheet and /render-html workflow used in academic-research production, and the skills/ and tools/ directories sit at the top level of the repository. If you are evaluating agent-generated long-form technical content, the artifacts are here to inspect. ## How the /interview-cheatsheet and /render-html pipeline produces a single-file page The data flow is a two-skill chain. /interview-cheatsheet drafts the content and /render-html turns it into the HTML you read. The README names the outputs directly: the cheat sheet HTML files live under docs/tutorials/, and the runnable code that accompanies them lives under docs/tutorials/code/. The newest entry in the What's New list, Modern Diffusion Post-Training, points at docs/tutorials/code/diffusion_online_rl.py, which the README says contains analytic checks of marginal preservation, the NFT update sign, and DGPO's balanced weights. That is a concrete example of the layout: prose page plus a companion script in a parallel directory. The rendering contract is stated in the README's HTML section. MathJax renders LaTeX formulas as text rather than screenshots, so they stay scalable, copyable and selectable. highlight.js colors the PyTorch blocks. The layout is responsive, there is a sticky table of contents, and the whole document is a single-file HTML that can be downloaded once and read offline with no backend. Those choices explain why the project ships HTML instead of a docs site: the artifact is the deliverable, and a link opens the same way on a phone, a tablet or a laptop. The generation process is described in the release notes as reviewed rather than one-shot. The 2026-09-10 entry says the sheet was design-reviewed before drafting with 64 guardrails, then went through three review rounds, all with gpt-6-astra xhigh. The README does not publish the guardrail list, the review rubric or the model configuration files, so the quality claim rests on the output rather than on a reproducible recipe. ## Installing ARIS-in-AI-Offer and rendering your first cheat sheet There is no package on PyPI and no release artifact published for this repository. The repository is the distribution channel, and the README points readers at the HTML files rather than at an install command. The practical first step is to clone the repository and look at what is already rendered. ```bash git clone https://github.com/wanshuiyin/ARIS-in-AI-Offer.git cd ARIS-in-AI-Offer ls docs/tutorials ``` The listing should show the rendered cheat sheet HTML files, including the Diffusion Foundations page the README uses as its preview and the newer modern_diffusion_post_training_tutorial.html. Opening any of them in a browser is the intended reading experience: MathJax and highlight.js load, formulas stay selectable, and the sticky table of contents tracks your position. Because the pages are single-file, you can copy one to a phone or an e-reader and read it without a network. The generated code samples are worth running separately. The README names one script explicitly: ```bash python docs/tutorials/code/diffusion_online_rl.py ``` According to the README, that script performs analytic checks of marginal preservation, the NFT update sign, and DGPO's balanced weights, so the expected output is a set of numerical assertions rather than a training run. Treat it as a verification of the sheet's formulas, not as a benchmark. If you want to regenerate content instead of reading it, the skills/ directory is where the ARIS skills live, and the README attributes generation to the /interview-cheatsheet and /render-html skills from the separate ARIS main repository. This repository does not document the invocation syntax for those skills, so check the skills directory contents before assuming a command name. ## MathJax, highlight.js and the offline single-file trade-off Shipping HTML instead of Markdown has costs the README does not discuss. MathJax and highlight.js are external JavaScript dependencies, and the README describes the output as single-file HTML that reads offline. Those two statements sit in tension unless the libraries are inlined or cached, and the README does not say which. A reader who downloads a page and opens it on a plane may find formulas rendered as raw LaTeX if the CDN is unreachable. That is the first thing to verify before trusting the offline claim. The second cost is searchability. A single-file HTML page with a sticky table of contents is pleasant to scroll, but it is harder to grep across a collection than a directory of Markdown files. If your study habit is full-text search across all 34 sheets, the HTML form works against you. The repository does keep README_CN.md and README.md at the top level, and the docs/ tree preserves the source layout, so the content is not locked away, but the reading artifact and the searchable artifact are not the same file. ## The ARIS-Homepage generator and its DBLP fact-checking step The repository bundles a second tool with a different purpose. The README describes ARIS-Homepage as the same /render-html workflow turning a CV into a fact-checked academic homepage, and the repository description calls it a CV to DBLP fact-checked academic homepage generator. The live demo is linked at wanshuiyin.github.io, and the preview image shows a header and bio, an ARIS Featured section with a hero SVG floated right, and publications grouped by topic with thumbnails. The interesting design decision is the fact-checking step. A CV is a self-reported document, and publication lists drift. Routing the publication entries through DBLP before rendering is a narrow, verifiable check rather than a general proofreading pass. It will not catch a misstated role or an inflated description of your own contribution, and the README does not describe what happens when a paper is missing from DBLP, which is common for preprints, workshop papers and non-CS venues. Treat the check as a filter for bibliographic metadata, not as a correctness guarantee for the page. ## Where the generated sheets stop being the right tool The clearest limitation is language. The repository is labelled bilingual, but the README is blunt that each cheat sheet is a long-form Chinese tutorial. The English side of the bilingual claim appears to be the repository-level documentation, README.md alongside README_CN.md, not a parallel English rendering of all 34 sheets. If you cannot read Chinese, you get the index and the framing, not the derivations or the 25 questions per topic. That is a hard boundary, not a soft one. The second limitation is maintenance shape. There are no releases in this repository, and it is a content collection rather than a versioned library. The What's New list shows dated entries, the most recent being 2026-09-10, and the last push to the repository was 2026-09-10. Content updates arrive as new sheets and revisions, not as semantic versions you can pin. If your workflow depends on a dependency that does not change under you, this is the wrong shape. The third is the review process. The README describes design review with 64 guardrails and three review rounds for the newest sheet, but the guardrails, the review prompts and the model configuration are not published. You cannot rerun the review or audit it. For interview preparation that is tolerable. For citing the sheets as a technical reference, the missing provenance matters. ## ARIS-in-AI-Offer against a general interview question bank The obvious alternative is a broad interview question bank such as the ones that collect ML and LLM questions across many companies in one searchable site. The difference is depth and format. A general question bank optimises for breadth and quick lookup: short questions, short answers, often organised by company or by role. ARIS-in-AI-Offer optimises for a single topic at a time, with derivations, from-scratch PyTorch and a stratified question set attached to the same page. You get one coherent document per topic instead of fragments across a search index. The second alternative is a textbook or a survey paper on the same subject. Those carry authorial accountability and stable citations, which the generated sheets do not. What the sheets add is the interview framing: the L1, L2 and L3 stratification tells you which questions are screening-level and which are top-tier-lab-level, and the from-scratch code is chosen to be short enough to reproduce on a whiteboard. A textbook will not make that distinction for you. ## Licence and upgrade cost for the cheat sheet collection The repository is MIT licensed, and the LICENSE file sits at the top level. MIT is permissive: reuse, modification and redistribution are allowed with the copyright notice and permission notice retained. Two caveats are worth naming without turning into legal advice. First, the repository bundles third-party assets under assets/, including preview images and a logo, and their provenance and terms are not stated, so the MIT grant may not cover everything in the tree. Second, the cheat sheet content is generated text that discusses published research; the MIT licence covers the repository's own files, not any rights in the underlying papers being summarised. The upgrade cost is low in the mechanical sense and non-trivial in the editorial sense. There is nothing to compile and no dependency graph to resolve, so pulling a newer version is a git pull. What you cannot do is diff meaningfully: a revised sheet can change a derivation or a code block without a changelog entry beyond the dated What's New bullet. If you have annotated a downloaded HTML file, your annotations will not survive the next revision. Keeping your own notes outside the repository is the only way to avoid that. ## Conclusion Adopt it if you are preparing for Chinese AI campus recruiting and want formula derivations, runnable PyTorch and stratified question banks in one offline HTML file, or if you want to study how a skills-based agent workflow produces long technical documents. Do not adopt it if you need a maintained, versioned package with releases and a test suite, or if you cannot read Chinese: the cheat sheets are long-form Chinese tutorials even though the repository is labelled bilingual. Before relying on the generated pages, verify that the ARIS skills directory in this repository is the version you intend to run, check the LICENSE file and the third-party assets under assets/ for their own terms, and open one rendered HTML file end to end to confirm the MathJax and highlight.js assets resolve offline. ## FAQ ### What is ARIS-in-AI-Offer? It is a bilingual collection of ML, LLM, multimodal, diffusion, agent and generative-model interview cheat sheets aimed at Chinese AI campus recruiting, generated by the ARIS /interview-cheatsheet and /render-html workflow. The README describes 34 first-party cheat sheets across 7 categories plus one community-contributed category, each with formula derivations, from-scratch PyTorch code and 25 stratified interview questions. ### How do I install ARIS-in-AI-Offer? There is no published package or release for this repository. The repository itself is the distribution: clone it, then open the rendered pages under docs/tutorials/ in a browser, since the README states the output is single-file HTML that reads offline with no backend. ### Is ARIS-in-AI-Offer available in English? The repository is described as bilingual, with README.md and README_CN.md at the top level, but the README states that each cheat sheet is a long-form Chinese tutorial. The English side applies to the repository documentation rather than to a parallel English rendering of all 34 sheets. ### What is the ARIS-Homepage generator in this repository? It is the same /render-html workflow applied to a CV, producing a fact-checked academic homepage. The repository description calls it a CV to DBLP fact-checked academic homepage generator, and the README links a live demo at wanshuiyin.github.io. ### What licence does ARIS-in-AI-Offer use? The repository is MIT licensed, with the LICENSE file at the top level. The terms and provenance of the bundled third-party assets under assets/ are not stated, so the MIT grant may not cover every file in the tree. ## Sources - [Issues](https://github.com/wanshuiyin/ARIS-in-AI-Offer/issues) - [License: MIT](https://github.com/wanshuiyin/ARIS-in-AI-Offer/blob/main/LICENSE) - [README](https://github.com/wanshuiyin/ARIS-in-AI-Offer/blob/main/README.md) - [wanshuiyin/ARIS-in-AI-Offer on GitHub](https://github.com/wanshuiyin/ARIS-in-AI-Offer) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/wanshuiyin-aris-in-ai-offer