Tags
Writing, research, projects, and photos around “Technology”.
Using multi-backend model access as the setting, this article explains how an AI Gateway handles resource scheduling, capacity control, failure switching, distributed state, usage accounting, and enterprise cost governance.
Short
16 min read
Part of the column “Technical Systems” · Chapter 9
Content, search, commerce, and growth systems repeatedly face the same challenge across multiple countries, industries, and scenarios: how to reuse stable capabilities while keeping business expressions distinct. How should boundaries be drawn between configuration, protocols, components, and orchestration?
Short
12 min read
Part of the column “Technical Systems” · Chapter 8
Using a multi-layer execution service as an example, this article breaks down the multiple levels of locality across request identity, edge routing, execution resources, and backend caches, then presents reusable approaches to routing, invalidation, failover, and observability.
Short
13 min read
Part of the column “Technical Systems” · Chapter 7
Starting from the fault structure of a multi-node execution system, this article reviews how requests, state, storage, performance, and failures gradually cross boundaries, and discusses how the next architecture should be reorganized.
Short
19 min read
Part of the column “Technical Systems” · Chapter 6
Starting from the observability problems of a small API service, this article distills the minimum implementation of request IDs, distributed Traces, service discovery, health monitoring, log queries, reliability, and infrastructure governance.
Case
15 min read
Six independently published skins, aggregate bundle loading, avatar identity convergence, partial-install fallback, and a clean uninstall boundary for DSH Web.
View projectA practical account of developing and debugging ChatLab, Agent Suite, and Quota Router, covering DSH plugin mechanics, common preview-version stability problems, and the boundaries that are easiest to get wrong when building plugins yourself.
Case
12 min read
DSH has been out only a few days, and I've already installed nine plugins on it. Notes on what each one fills in, how it works, and which one is the most useful.
Short
5 min read
Using a base-plus-skin architecture to give the DSH Web GUI a switchable, uninstallable Feishu-style chat skin that touches no existing plugin: workspaces become project groups, sessions become contacts, and the chat becomes bubbles.
View projectFrom a temporary SSH direct connection, to accommodating phones and family, and finally converging on sing-box as a maintainable return-to-China network setup.
Case
10 min read
A position paper based on the author's experience driving AI infrastructure and pilots in a real engineering organization: it proposes four kinds of boundaries—context, responsibility, authorization, and measurement—and offers a falsifiable pilot protocol plus a minimal harness as a demonstration of putting them into practice.
9 min read
Read summaryA research-design and protocol paper: reframing data measurement from what belongs on the dashboard into a recomputable, reviewable, decision-supporting collaboration protocol, with minimal mechanisms for definitions, measurement units, metric tiering, dictionaries, and retrospectives.
11 min read
Read summaryA research synthesis and position paper: defining engineering productivity as a continuously shortening "propose change — get trustworthy feedback — correct safely" loop, with metrics, a default path, and a falsifiable pilot protocol.
9 min read
Read summaryA system design paper: modeling long-form fiction collaboration as recoverable, auditable, author-approved state transitions, and proposing an evaluation scheme that does not mistake mechanical scores for literary value.
Read summaryTechnical planning is not a project list. At its core it is two things: understanding the business you serve (business analysis) and understanding the gap and level against competitors (competitive analysis), then linking them into a causal chain of goals, trade-offs, and execution.
Short
16 min read
Part of the column “Technical Systems” · Chapter 1
A multi-region system can neither cram every difference into a single core nor rebuild everything for each region. The key is to identify stable boundaries, preserve extension points, and clarify module ownership.
Short
3 min read
Part of the column “Technical Systems” · Chapter 4
Every role owns its own delivery, and everyone shares accountability for production outcomes. A quality loop depends on monitoring, fast containment, precise fixes, and retrospectives.
Short
2 min read
Good standards don't turn every step into an approval; they make the key handoff points — where distortion, rework, and accidents are most likely — visible, discussable, and reusable.
Short
2 min read
Part of the column “Technical Systems” · Chapter 3
Code decay rarely comes from any single moment of bad writing; it comes from local changes repeatedly bypassing shared boundaries. What truly needs maintaining is the consistency between design and collaboration.
Short
2 min read
Part of the column “Technical Systems” · Chapter 2
From daily and weekly reports to problem retrospectives: how to maintain baselines, record changes, judge impact, and turn data conclusions into verifiable action items.
Short
4 min read
Part of the column “Data Metrics Guide” · Chapter 5
A publicly reusable metric dictionary: from requests and users to tasks, explaining how availability, error, latency, performance, and feedback data should be defined, combined, and interpreted.
Short
13 min read
Part of the column “Data Metrics Guide” · Chapter 2
Metrics are not numbers on a report; they are the shared language a team uses to describe the same thing. Only after defining the object, event, denominator, and time can data participate in decisions.
Short
6 min read
Part of the column “Data Metrics Guide” · Chapter 1
The value of an engineering retrospective is not in retelling what happened or blaming individuals, but in turning an experience into improvements that can be verified, maintained, and reused.
Short
6 min read
Turn technical sharing from a regular talk into a knowledge-collaboration system that connects topic selection, research, discussion, and consolidation.
Short
5 min read
An entry page is not a pile of links, but the shared memory a constantly changing project leaves for its team: it helps readers enter by task, judge whether information is still valid, and find the people responsible for maintaining it.
Short
5 min read
Part of the column “Documentation & Knowledge” · Chapter 2
An old signpost written for frontend beginners: information gathering, study logs, algorithm practice, interviews, and internships are all just the beginning.
Method
2 min read
An index of interview and engineering fundamentals compiled after the spring 2018 job search; keeping it as a reminder that interview questions are never just interview questions.
Method
1 min read