Note: public references should use website links. Unpublished drafts keep titles only and should not expose internal repository paths.
This corpus organizes Changkun Ou's long-term knowledge domains, recurring questions, and writing tendencies by theme. The emphasis is on durable methods, judgments, and boundaries rather than on a list of experiences. The source material goes beyond blog posts and includes education, research, work, teaching, and open-source experience, as well as GitHub repositories that already function as courses, paper artifacts, tutorials, translations, reading lists, or knowledge infrastructure.
It is closer to a layered knowledge map than to an archive. The foundation layer covers mathematics, systems, and language judgment. The middle layer covers research formation and human-in-the-loop problems. The upper layer covers teaching, open-source work, and writing as public output. The outer layer preserves the background and edge zones provided by cross-cultural life and long-tail writing.
- The reference material spans public posts, public notes, topic pages, tutorial pages, project pages, and a small number of unpublished drafts from 2009 to 2026.
- Formal education, research, work, teaching, and open-source experience are used to calibrate the timeline, professional identity, and domain boundaries.
- Research material is anchored mainly in the public research record: research pages, papers, theses, defenses, and talks. Unpublished research plans are used only to clarify recent directions, not to claim finished results.
- Public GitHub repositories are also read as core source material, especially high-density knowledge repositories such as course repos, research artifacts, books, tutorials, translation catalogs, reading lists, and thesis templates.
- Images, static assets, tag pages, and site configuration files are outside the scope.
- Earlier imported QQ Space texts contain many duplicate drafts and short notes. Priority is given to texts that are complete enough to reflect stable concerns.
- The starting point is mathematics, especially set theory, axiomatization, order, measure, proof structure, and the question of why a definition is set up the way it is.
- After moving into computer science, the focus turns to why systems hold together: runtime, kernels, memory, performance measurement, source design, infrastructure, and engineering constraints, without being tied to one framework.
- AI is approached through statistical learning theory, HCI, human computation, social systems, agency, and self-reference, rather than through model optimism alone.
- The formal research line centers on how humans enter decision, optimization, and feedback loops, across behavior understanding, human-in-the-loop optimization, adaptive VR/AR, crowd + machine learning, and agent oversight.
- Several high-investment case domains recur across research, teaching, and engineering, but they function mainly as carriers of method and problem consciousness.
- These interests later take shape as HCI and human-machine-loop research, LMU teaching, cross-platform product development, enterprise AI platforms, and revenue optimization systems.
- Recent public notes show the AI focus moving toward agent workflows, human-in-the-loop oversight, responsibility chains, confirmation fatigue, preference modeling, and self-referential mechanisms.
- The knowledge is expressed not only through long-form essays, but also through course repositories, tutorial books, research artifacts, project archives, and interaction prototypes.
- Open source and public output form a durable part of the professional identity: the Go and Fyne communities,
golang.design, TalkGo, translation work, and public tutorials make up another stable line. - The long migration from Chengdu to Munich, together with switching across Chinese, English, and German, gradually adds cross-cultural and institutional comparison to later writing.
- In recent years the cultural center of gravity has also shifted from "how to build" toward "why to build, for whom, and how a structure remains stable over time."
- Mathematics and Learning Theory
- Systems and Engineering
- Programming Languages, Platforms, and Product Judgment
This layer captures the deepest recurring tendencies: axiomatization, mechanism-first thinking, and a sense for product boundaries.
This layer covers the central formal problem space across research and industry. One page focuses on how research training, methods, and major clusters were formed. The other focuses on how those lines converge on human-machine loops, preference, responsibility chains, and organizational systems.
- Teaching and Course Design
- Open Source, Talks, and Public Output
- Writing, Research, and the Cultural World
This layer shows how knowledge is reorganized into courses, projects, papers, books, templates, communities, and public expression. One page emphasizes teaching and training structures, one public repositories and knowledge infrastructure, and one writing method, research judgment, and cultural orientation.
This layer is not the trunk itself, but it explains why the trunk changed in the way it did. It also keeps materials that genuinely shaped the process without fitting cleanly into the core. One page preserves stable life background and media environment. The other keeps long-tail material that has not settled into the main line.
- Axiomatization: ask for the minimal premises of a system, the construction of objects, and the necessity behind definitions.
- Structural thinking: start from whole structures, constraints, and boundaries rather than from loose collections of tricks.
- Mechanism first: when facing performance, theory, or engineering questions, trace the underlying mechanism before talking about surface practice.
- Depth first: over time, place more weight on depth built through long practice and less on breadth for display.
- Cross-domain framing: mathematics, engineering, social choice, philosophy, and personal experience often enter the same chain of questions.
- 2009-2013: intensive self-study in mathematics, then entry into C/C++, algorithms, OpenCV, Linux, and foundational coursework.
- 2014-2018: a turn from computer vision and machine learning toward statistical learning theory, HCI, human computation, and systematic writing, together with a research-training shift in Germany.
- 2018-2023: the LMU period, where web behavior research, human-machine loops, human-in-the-loop optimization, adaptive VR/AR, course design, case-domain work, Go runtime study, and public output advance in parallel.
- 2022-2026: enterprise engineering and AI platform work, where revenue optimization, agent workflows, knowledge management, preference modeling, and organizational AI transformation become production contexts for earlier research interests.
Viewed through the four layers above, this line becomes clearer: the foundation layer forms first, then converges into the research layer, then keeps growing outward into the output layer, while the background layer continues to shift the center of gravity from the outside.
- Durable knowledge lines that recur over many years, rather than a running list of technologies once used.
- Questions that remain stable across time, rather than short-cycle judgments driven by hype.
- Thinking habits, domain boundaries, and cultural orientation reconstructed from long-term writing, projects, and formal experience, without self-packaging.
- Each topic page focuses on relatively stable knowledge domains.
- Long-Tail Topics and Unstable Writing keeps stage-specific judgments, one-off technical experiments, cultural commentary, and unfinished drafts to preserve a fuller writing map.
- Terminology maintains cross-page Chinese/English term pairs and preferred default forms.
- Agent Profile provides a self-contained system prompt that distills the methodology, knowledge domains, values, and problem-solving approach from the entire corpus into an operational profile for agents.
- Chinese mirror pages are stored under
zh/and linked from the language badge at the top of each page.