DiVo Learning Assistant: Let Every Piece of Knowledge Echo

TL,DR (too long, don't need to read)

Actually this update, from V0.3 to V0.5, was mainly about packaging a standalone executable EXE for Windows. Moving from a pure web app to a desktop app across two platforms.

Linux is easy to package, but with too few users we didn't bother. Same for macOS.

Of course, we're still testing packaging an Android APK, but there's a serious obstacle left unsolved.

We'd also like to package an iOS app. If enough people want it and support me in getting Apple App Store certification.

Linux creator Linus has a famous saying: talk is cheap, show me the code. Although in the AI era many feel code is cheap and you just show the AI the talk (prompt), in reality an AI coding project that independently develops its own app with complex business logic, complex data structures, and complex network architecture requiring elaborate engineering solutions cannot be done by lip service alone.

Below is a brief introduction to DiVoLA's technical highlights

Of course, on the technical side there is much necessary knowledge and experience for efficient work and effective architecture design and content modeling—for example, the Linux OS, the TCP/IP protocol and network architecture... but these can all be counted as unified within Drupal's learning curve. To use Drupal skillfully to architect, deploy, and optimize a site for performance and UX, you gradually learn all this. But now it's even better—you don't really need to know that much; just ask the AI and let the AI engineers do the work.

Then, as AI can replace us in everything—front-end design, programming, back-end design, server ops, network ops, documentation, general office work—is there still a need to learn? We'll write a "Learning for the AI Era" later based on the feel of this project's development. We're also preparing the course series 《LLM Based Generative AI USE》. No more on that here.

This is not just line after line of code.

This is a lamp we personally lit, in this noisy digital age, for every soul that longs for quiet reading and thinking.

Actually that sounds a bit hollow. What I mean is: most of the exam points that let most test-takers score 85% are right there on the syllabus.

A teacher who sets out-of-syllabus papers is an idiot; don't stoop to their level. Why stoop to an idiot's level?

Besides, as I said, an exam isn't about solving every problem, but about scoring as many points as possible in the limited time.

So when I created the "Straight-A Student Handbook" the year before last, one very important point—a super important point—was

Back to the textbook! Back to the textbook! Back to the textbook!

But when I tutored a kid last summer, I found a problem: the people writing textbooks nowadays aren't decent either. The textbooks they write have poor continuity, flashy layouts, and fragmented content. I don't get why they'd make a textbook look like a quack's plaster. And some have no basic Chinese writing ability—can't even write a complete sentence, often missing the main clause. I see this often when guiding middle- and high-school science study.

So today's kids have it worse than we did; even such vital learning material as textbooks has much worse quality.

But after all, the textbook embodies the syllabus, and the syllabus throttles the exam outline. So to score high, master the textbook's knowledge points thoroughly, form a system, understand—don't rote-memorize—and you'll be at ease and handle the exam with room to spare. Like when I won the Hunan provincial biology olympiad first prize in middle school: no tutoring, no past papers, just two textbooks plus my own rural-life observations.

But the reality is the opposite: for most students, the exam that should be calm and composed ends up rushed and flustered.

Why? I think the key problem is that many test-takers aren't familiar enough with the textbook. What does "read a book a hundred times and its meaning appears" mean? What does "cram at the last minute" mean?

Look at those "postgrad-prep lifers" universities, where "Biochemistry" is taught twice during undergrad. Those students' written exam scores clearly have the edge.

Of course there's another big problem: many teachers themselves aren't familiar with the textbook and can't explain clearly. Unclear teaching, unfamiliar textbook, no own understanding or knowledge system—either they read the book aloud, or they just replay the PPT. Some blindly pile on intensity and all sorts of extracurricular references. Students haven't even grasped the textbook—why all these flashy crooked tricks?

Alright, enough negativity; let's repeat the point of points

To score high, best to know the textbook thoroughly!

So that was my original idea for developing this DiVo Learning Assistant.

On content and features:

Based on the official textbook, split into units by lesson, then use an AI voice engine to generate read-aloud audio from the text. Students can listen to the lesson, record themselves reading along, read the textbook lesson content, and take notes through our app.

Those are the main features. Simple, with an e-ink-style interface suited to focused reading and listening.

In a world where information surges like a tide yet attention grows scarce, DiVo Learning Assistant was born. It should not be cold screen-swiping, but an immersive journey for eyes, ears, and heart together.

I. The Core Foundation: Standing on the Shoulders of Giants

DiVo's technology choices reflect an obsession with ultimate engineering aesthetics.

The specific tech stack is listed in the image above. Thanks to so many open-source technologies and so many AI-coding IDEs and CLIs.

So by leveraging AI's superstrong coding ability—though I wrote not a single line of code, communicating purely in natural language with AI—we still completed very solid programming and development work.

Metric dimension	Core data	Notes
Total lines of code	20,403 lines	Pure source stats, excluding node_modules and build artifacts
Total source files	300+	Includes TypeScript (.ts/.tsx) and JavaScript core logic
React component matrix	80+	Covers UI base components, business-logic components, and large page layouts
Core API routes	1+ (sync-audio)	Core interface supporting high-concurrency audio sync and content pipelines
Architecture flexibility	Multi-dimensional Hooks & Utils	Complete custom Hooks logic and cross-platform utility function library

II. Hear the Voice of Knowledge: The Full Data Dashboard

Dimension	Data metric	In-depth analysis
Total lessons	2,088	Covers most knowledge nodes from elementary, middle, and high school; some textbook content we couldn't obtain, so be it
Total audio segments	5,847	For long texts we use smart segmentation to ensure zero-latency loading
Total real-time playback duration	580+ hours	A conservative estimate. Even listening non-stop, it would take 24 days to finish
Physical storage size	3.92 GB	Every byte compressed with 32kbps/48kbps high-fidelity voice compression
AI voice engine	MicroSoft vibevoice	We chose a gentle female voice as the main timbre for generating audio files

III. Craftsmanship: The Warmth Wrapped in Code

1. E-Ink Optimization for Eye Rest

We developed E-ink Flash Overlay specifically for users studying on E-ink devices. Simulating the traditional "full black refresh" on route changes is not just for adaptation, but a promise of focused, eye-friendly reading.

2. Shadowing Recorder

We built a local recording museum based on IndexedDB, along with local study notes. Learners can listen to a segment, record a segment, no network needed, no data worry—your effort is saved at the browser's edge node, echoing anytime.

3. Feynman Learning Method (Learning Matrix)

DiVo is never just "read and done." The UI deeply integrates a five-layer cognitive model:

Initial exposure → Memorized → Understood → Applicable → Teachable
Every click to mark is one step up the cognitive pyramid.

IV. Industrial-Grade "Invisible" Infrastructure

Global delivery: Relying on CloudFront CDN and R2 edge storage, even if you're in a remote place, knowledge flows as clear and smooth as a spring.

Security barrier: Electron's contextBridge isolation and .env.local production rules keep your account and API keys as safe as a vault.

Ultimate ops: Through sync-all-audio.ts and warm-up-cdn.ts, every update of new content is gently laid down in the background automatically.

V. Vision: Know the Textbook Better, Score Higher

What we write is not code, but hope.
What we count is not bytes, but harvest.

DiVo Learning Assistant Hear the voice of knowledge.
Keep the boundaries of learning always in your grasp.

🌐 Try online: divola.wangjueju.cn 🔗 APP download: click the "Download APP" button on the home screen to enter the download page

PHP 8.5 + FrankenPHP Worker Mode + Native PG/Redis // All by AI

Why not just rewrite Drupal in Rust?

DiVoLA V0.5.0 Released

Upgrading PHP to 8.5 for Drupal 11.3

Uncle Ju AI Assistant Platform — Complete Technical Analysis Report for the DiVoAI Project

Deploy the vector database first

Blindly Applying AI to Everything Will Only Harm You

Testing the Second AI Writing Implementation

Implementing Multiple Auth Methods for MCP

Turning a Drupal Instance into an MCP Server

1. Turning Drupal into a Knowledge Base: RAG-Ready Drupal

On the Positioning of Drupal Development

Added a Content-to-Markdown Text Format Feature

Google Fonts, Composer Libraries, and Various JS CDNs Are All Blocked

Testing the Upgraded System

A Rant About Drupal's Management!

Enabling HTTP/3 and QUIC Support for nginx

drush-bash: drush: command not found

Enabling Remote Access for Redis

给我的drupal网站装了一个朗读模块

nextcloud or drupal?

关于drupal位于nginx proxy后使用pdf.js出现file origin does not match viewer's的解决方案

好像很多链接出错了，还得逐一修正

caches make your website fly

trusted host一定要记得设置

把editor放到了前端 极大地促进了发表量

Multilingual Drupal and Automatic Translation (logged Mon, 09/23/2019 - 11:06)

Publish a Short Post If You Want to Publish a Short Post

Still Need to Restore and Import Some Old Posts

With This, Writing Becomes Fun.

测试一个邮件发blog的服务

把editor放到了前端极大地促进了发表量