# A-Znk Blog > Master's student at USTC · AI safety and large language model systems Author: A-Znk (academic name: Qianli Ma), Master's researcher in AI Safety & LLM Systems at USTC (ORCID: https://orcid.org/0009-0005-2060-4172). Full clean-text corpus: https://a-znk.com/en/llms-full.txt · Chinese index: https://a-znk.com/llms.txt ## Author - [A-Znk](https://a-znk.com/en/author): Master's student at USTC · AI safety and large language model systems · Publications as Qianli Ma ## Papers - [DUAP: Disentanglement-Based Universal Adversarial Perturbations for Robust Multilingual Speech Privacy Protection](https://a-znk.com/en/papers/duap-multilingual-speech-privacy): 2026 · IEEE Transactions on Information Forensics and Security · published · DOI: 10.1109/TIFS.2026.3671687 · Language feature disentanglement and gradient optimization produce transferable perturbations that reduce Whisper’s ability to transcribe sensitive speech. ([Markdown](https://a-znk.com/en/papers/duap-multilingual-speech-privacy.md)) - [Ratiometric Determination and Discrimination of Oxicams via Dual-Excitation Carbon Dots Assisted by Machine Learning](https://a-znk.com/en/papers/oxicams-carbon-dots-machine-learning): 2025 · Analytical Chemistry · published · DOI: 10.1021/acs.analchem.5c03226 · Combining dual-excitation fluorescence ratios, carbon-dot sensing, and machine learning for sensitive quantification and discrimination of oxicam drugs. ([Markdown](https://a-znk.com/en/papers/oxicams-carbon-dots-machine-learning.md)) ## Posts - [A-Znk Blog: Design Manual](https://a-znk.com/en/posts/myblog-design-manual): 2026-04-11 · How the code, runtime boundaries, and test contracts organize topic-first discovery, reading-first articles, and restrained interaction into a shared design language. ([Markdown](https://a-znk.com/en/posts/myblog-design-manual.md)) - [Ant AI Coding Assessment: A Retrospective](https://a-znk.com/en/posts/ant-ai-coding-review): 2026-03-19 · Preparation, execution, pitfalls, and a timeline of an Ant AI coding assessment. ([Markdown](https://a-znk.com/en/posts/ant-ai-coding-review.md)) - [A Minimal Implementation of Paragraph Anchors and Comments](https://a-znk.com/en/posts/paragraph-anchor-design): 2026-02-20 · Inject stable anchors with rehype and connect them to Supabase for extensible paragraph-level comments. ([Markdown](https://a-znk.com/en/posts/paragraph-anchor-design.md)) - [Why Topic Pages Should Be a Blog’s Main Entry Point](https://a-znk.com/en/posts/why-topic-first): 2026-02-20 · The benefits of moving from a chronological feed to a topic map, with a practical way to get started. ([Markdown](https://a-znk.com/en/posts/why-topic-first.md)) ## Topics - [Topic-Based Knowledge Networks](https://a-znk.com/en/topics/knowledge-network): Organizing knowledge by topic and treating articles as reusable nodes. - [Paragraph-Level Comments](https://a-znk.com/en/topics/paragraph-review): Using the paragraph as the smallest unit of discussion, so feedback attaches precisely to the text. - [Agent Engineering](https://a-znk.com/en/topics/agent-engineering): How prompts, rules, tools, tests, and task boards come together as executable agent systems. - [AI Coding](https://a-znk.com/en/topics/ai-coding): AI coding tools, workflows, constraints, and the experience of delivering real software. - [Assessment Retrospectives](https://a-znk.com/en/topics/exam-review): Strategies, mistakes, and lessons from technical assessments, system design problems, and timed development. ## Concepts - [Anchor ID](https://a-znk.com/en/concepts/anchor-id): A paragraph’s stable locator within an article, composed of a section slug and a paragraph number. - [Optimistic UI](https://a-znk.com/en/concepts/optimistic-ui): Update the interface before server confirmation to provide faster feedback on comment interactions. ## Optional - [Full Markdown Corpus (llms-full.txt)](https://a-znk.com/en/llms-full.txt): Single-file Markdown bundle of author profile, published papers, posts, topics, and concepts - [Chinese LLMs Index (中文索引)](https://a-znk.com/llms.txt): Authoritative Chinese source index - [Agent Work Observatory](https://a-znk.com/en/agents): Interactive instruments, public field-study evidence, and reviewed agent delivery records - [Wander · Knowledge Graph](https://a-znk.com/en/map): Topic-post-concept adjacency explorer and curated reading trails - [Knowledge Graph JSON](https://a-znk.com/knowledge-graph.json): Machine-readable nodes and edges connecting topics, posts, and concepts - [A-Znk Blog RSS](https://a-znk.com/en/rss.xml): RSS 2.0 feed of English articles