DEV Community: MinJian

I Built HSKKit: A Different Way to Learn HSK 3.0 Vocabulary

MinJian — Mon, 14 Sep 2026 02:43:33 +0000

Learning Chinese vocabulary looks simple at first.

You get a word list, memorize the words, and move on to the next level.

But while looking at how HSK learners actually study, I noticed a problem: having access to thousands of words is not the same as having a clear learning path.

A learner might have a vocabulary list, a flashcard app, a dictionary, a listening resource, and a mock test — but these tools often exist separately.

That was the problem I wanted to explore when building HSKKit.

The problem I wanted to solve

The new HSK 3.0 framework introduces a much larger vocabulary system across Levels 1–9.

For someone preparing for HSK, several questions appear very quickly:

What level should I start from?
Which vocabulary should I learn today?
When should I review a word?
How do I know which words I keep getting wrong?
How can I combine vocabulary and listening practice?
How do I avoid repeatedly studying words I already know?

A static vocabulary list doesn't really solve these problems.

So instead of building another directory of HSK words, I wanted to experiment with a more connected study workflow.

What I built

The result is HSKKit, a web-based learning platform organized around published HSK 3.0 vocabulary.

The current workflow connects several parts of studying:

Placement → Vocabulary → Listening → Practice → Review → Mistakes

Learners can first find a practical starting level and then move into vocabulary organized by HSK level or topic.

The platform currently includes HSK 3.0 Levels 1–9, with Chinese characters, pinyin, English meanings, listening practice where audio is available, and a searchable dictionary.

The interesting part: deciding what to review

One of the biggest design decisions was not simply what content to show, but when to bring content back.

If someone already remembers a word easily, showing it again tomorrow isn't very useful.

If they repeatedly forget another word, that word deserves more attention.

That's why I added a spaced-repetition-style review flow.

Instead of treating every vocabulary item equally, the study experience tries to bring words back when they actually need another review.

Mistakes are also kept in the learning loop rather than disappearing after a practice session.

Learning by situations instead of only levels

Another thing I wanted to experiment with was topic-based learning.

HSK levels are useful for exam preparation, but people don't experience a language as "Level 1 words" and "Level 2 words" in real life.

They experience situations.

For example:

getting ready in the morning
staying at a hotel
doing household chores
making plans
travelling
ordering food

So HSKKit also groups vocabulary into practical situations while keeping the words connected to their published HSK levels.

This creates two possible paths:

Exam-oriented: learn by HSK level.

Usage-oriented: learn vocabulary through real-world topics.

AI without replacing the source material

I also wanted to be careful about how AI is used.

For a language-learning product, simply asking an LLM to generate unlimited vocabulary or definitions introduces an obvious problem: the learner may not know whether the generated material actually belongs to the HSK standard.

So the core vocabulary catalog is source-controlled rather than freely generated by AI.

AI-assisted features can help around the learning experience, while the underlying HSK vocabulary remains tied to the published dataset.

This separation became an important product decision:

Use AI to improve the learning experience, not to invent the curriculum.

What I learned while building it

The biggest lesson for me was that an educational product doesn't necessarily need more content.

Sometimes it needs better connections between existing content.

A vocabulary database is useful.

A listening tool is useful.

A flashcard system is useful.

A placement test is useful.

But connecting them into one continuous path can make the product much easier to understand:

Where am I? → What should I learn? → Can I recognize it? → Can I hear it? → When should I review it?

That's the direction I'm continuing to explore.

I built the project here:

HSKKit: https://hskkit.com/

I'd especially be interested in feedback from developers who have worked on EdTech, language-learning products, spaced repetition, or AI-assisted learning.

What would you improve about this learning flow?

I Built an AI Career Test That Gives You Career Hypotheses, Not a Career Verdict

MinJian — Sat, 05 Sep 2026 03:42:39 +0000

Career questions have become harder to answer in the AI era.
It is no longer only about what someone is good at, what job title to pursue, or which skills to learn next. People are also asking which parts of their work AI can accelerate, whether a direction genuinely fits them or is simply popular online, and how they can test a career path before spending months on courses or making a major switch.
That is why I built AI Career Test, a free, roughly ten minute career self exploration tool. It is designed to turn personal signals into career directions to investigate and practical next steps.
It is not a professional assessment, career advice service, or a machine that can tell someone what job they should do.

The problem with a single career verdict
A result such as You should become a product designer can sound decisive, but it hides too much uncertainty.
A useful career exploration tool should look beyond a single personality label. It should help someone examine the combination of interests and work values, strengths and preferred ways of working, personality and collaboration signals, current skills, and AI era preferences such as learning, judgment, privacy awareness, and adaptability.
Even then, the result should be treated as a working hypothesis, not a prediction.
A test cannot know someone’s finances, location, qualifications, health, responsibilities, or the real hiring conditions in their market. It can help organize the signals already present and turn them into questions worth testing.

From assessment signals to an action plan
The free assessment contains 40 questions and uses a simple response scale rather than asking users to write a long biography.
The goal is not to label people. It is to help them reflect on patterns such as the kinds of problems they enjoy, their preferred work environment, whether they gain energy from analysis, creativity, collaboration, or execution, and what role AI should play in their work.
The result should lead to a better question than What is my perfect career. A more useful question is Which career directions are credible enough for me to validate next.

What a complete report should explain
The sample report on the site is intentionally fictional and marked as illustrative data. It demonstrates the kind of information I think a career report should provide.
A working hypothesis in plain language, including the strongest patterns and the uncertainty that remains.
Evidence linked career matches, with an explanation of why a path may fit rather than only a ranked list of job titles.
Trade offs and risks, showing what to test rather than only what looks attractive.
Skills and AI readiness, including learning, verification of AI output, privacy awareness, and judgment.
A 30, 60, and 90 day roadmap that creates momentum through small experiments, proof of work, and reflection.

Why test before you commit matters
A career match is more useful when it comes with a validation exercise.
Instead of telling someone to immediately pivot into a new field, a better next step could be to interview someone already doing that work, compare what the role involves during a normal week, build a small project resembling an important task in the role, collect evidence for one skill gap before buying another course, or rewrite one resume bullet around a real outcome instead of a generic responsibility.
The goal is to reduce blind commitment. Career exploration should move from vague self description to observable evidence. What work do I actually enjoy doing repeatedly. What can I show. What assumptions about a role have I tested.

Building around explainability
I wanted the product to make its limits visible. The site states that results are for self exploration, not professional assessment or career advice. The sample report also distinguishes between a recommendation, an uncertainty, and a next experiment.
AI can make reflection faster. It can help organize signals and reveal options people may not have considered. But it should not replace a person’s lived context, responsibility, or decision making.
The best outcome is not that the AI chose my career. The better outcome is that I have a clearer direction, understand why it may fit, and know the next small experiment to run.

Try the free assessment
If you want to try the free assessment or inspect the illustrative report structure, visit https://aicareertest.work/
I would also love to hear from developers and career changers. What would make you trust an AI assisted career exploration tool, and what would immediately make you distrust one?