Sirui's blog

Language: English | 简体中文

发布日期：2026-10-10

I built FLH’s read-only Knowledge Library to make curated Concepts searchable and traceable to the original learning question. The accompanying synthetic demo and acceptance work cover search semantics, source navigation, process ownership and reliable startup.

Engineering focus: Full-stack search, provenance and demo operations.

By this point the project could store questions, interpret them, extract units
and organize them into Concepts. What it lacked was a way to read that knowledge
back — “what do I know about the subjunctive, and where did it come from?” — and
a short demo that someone else could follow. Building the library was
straightforward. Making the demo honest took three more rounds.

语言： English | 简体中文

发布日期：2026-10-10

我构建 FLH 的只读 Knowledge Library，让整理后的 Concept 可以被搜索，并能追溯到原始学习问题。配套合成演示及验收覆盖搜索语义、来源跳转、进程身份与可靠启动。

工程重点：全栈检索、来源追踪和演示运维。

此时系统已能记录、解释、抽取和整理知识，却缺少把知识读回来的入口。建立页面并不难，让演示能可信地运行则经历了几轮修正。

Language: English | 简体中文

发布日期：2026-10-09

I added durable idempotency and recovery to Concept Review decisions so a lost HTTP response does not force the user to guess whether a write happened. The design joins a SQLite receipt, request identity and explicit browser recovery into one inspectable contract.

Engineering focus: API idempotency, transactions and frontend recovery.

Every decision in the Concept Review workbench is an HTTP POST. Usually the
browser sends it, the server commits it, and the response comes back. The
interesting case is when the response does not come back: the connection drops,
the laptop sleeps, a proxy times out. Did the write happen?

语言： English | 简体中文

发布日期：2026-10-09

我为 Concept Review 加入持久化幂等与恢复机制，让 HTTP 响应丢失时，用户无需猜测操作有没有写入。设计连接 SQLite 回执、请求身份与浏览器显式恢复流程。

工程重点：API 幂等、事务和前端恢复。

一次普通审阅是浏览器发 POST、服务器提交、响应返回。但连接断开、设备休眠或代理超时时，最后一步可能永远不到达。此时真正的问题是：写入发生了吗？

Language: English | 简体中文

发布日期：2026-10-08

I packaged the Go API and React workbench into a single Docker service, then hardened its backup, restore and browser-request boundaries. The work turned a development application into an operable personal release with explicit readiness and failure behavior.

Engineering focus: Release engineering, recovery and service boundaries.

A personal service is only useful if it starts the same way every time and its
data survives mistakes. This part of the project was less about features and
more about checking, from several angles, that the release did what its
documentation claimed.

语言： English | 简体中文

发布日期：2026-10-08

我把 Go API 和 React 工作台打包为同一 Docker 服务，随后完善备份、恢复和浏览器请求边界，把开发应用推进成有 readiness 和明确失败行为的个人使用版本。

工程重点：发布工程、恢复与服务边界。

个人服务的价值，取决于它能否稳定启动，以及数据能否从错误中恢复。这部分工作的重点是从多个方向检查发布行为是否符合文档。

Language: English | 简体中文

发布日期：2026-09-04

I designed the learning-record boundary in FLH so original questions, machine interpretations and human corrections have distinct ownership. The result is a versioned Go/SQLite workflow that preserves history, supports explicit correction and imports repeated captures without duplicate records.

Engineering focus: Data modeling, provenance and transactional ingestion.

French Learning Hub started from a narrow goal: questions asked while learning
French should not disappear inside individual chat conversations. That sounds
like a note-taking problem. It turned into a data-modelling problem as soon as
machine interpretation entered the picture.

语言： English | 简体中文

发布日期：2026-09-04

我为 FLH 设计了学习记录的数据边界，把原始问题、机器解释和人工纠正分开管理。Go / SQLite 工作流因此可以保留历史、接受显式纠正，并在重复导入 capture 时避免创建重复记录。

工程重点：数据建模、来源追踪与事务式导入。

FLH 最初要解决的问题很简单：法语学习中的问题不应该消失在一段段聊天里。引入机器解释后，原本的笔记问题变成了数据建模问题。

Language: English | 简体中文

发布日期：2026-09-03

I built the Concept and annotation model behind FLH: immutable extracted units remain evidence, while durable Concepts carry identity and current memberships. The same model supports a human review workbench, versioned dataset exports and comparable retrieval experiments.

Engineering focus: Domain modeling, review workflows and retrieval evaluation.

Once learning records were safely versioned, the next step was to turn them into
something reviewable: the individual ideas a learner actually met. This is where
the project’s most important distinction appeared. A piece of evidence that an
idea came up is not the same thing as the idea itself.

语言： English | 简体中文

发布日期：2026-09-03

我构建了 FLH 的 Concept 与标注模型：抽取结果保留为不可变证据，Concept 承担持久身份，当前 membership 表达有效归属。模型同时支持人工审阅、版本化数据集导出和可比较的检索实验。

工程重点：领域建模、审阅工作流与检索评估。

记录能够安全保存后，下一个问题是如何把学习中出现的想法变成可审阅对象。一个想法出现过的证据，并不等于这个想法本身。