Speed became part of the signal
Results pair accuracy with FAST, NORMAL, HESITANT, and REVIEW classifications, average effective recognition time, and words needing the most review.
Educational product case study
Letter World
A word-recognition platform built around a simple question: how can repeated practice become clearer, more focused, and more useful to students?
V2.0Current live release
18Classmates using it daily across five consecutive school days
2+Rounds per student per day during that window
10Students who specifically endorsed the ten-word format in qualitative feedback
30-second overview
The product, the role, and the evidence so far.
- Problem
- Students can know a translation yet still struggle to recognize a word quickly in context.
- The product
- A focused practice loop supported by a learning client, admin workflow, application API, and validation pipeline.
- My role
- Founder and Product Lead: product direction, learning goals, implementation coordination, version testing, and feedback-driven iteration.
- Evidence so far
- Eighteen classmates used V2.0 every day across five consecutive school days, completing at least two rounds per person per day. Ten specifically endorsed the ten-word format in qualitative feedback.
01 / The learning problem
Recognition must become fast enough to use.
Students may know the meaning of a word but still fail to recognize it quickly enough under time pressure.
Letter World began with a practical gap between knowing a translation and recognizing a word quickly in context. The product turns that gap into a focused practice loop rather than treating the game layer as the learning outcome.
Current release
What V2 Changed
Recognition speed, structured practice, and more useful feedback became part of the product loop.
Round length became selectable
Learners can choose 10, 15, 20, 25, or 30 words. Health scales with the selected round length.
Difficulty became structured
Generation targets 25% easy, 50% medium, and 25% difficult questions using whole-question quotas. Available content can require adjacent-difficulty substitutions; a question does not repeat within one round.
Later practice reflects hesitation
Incorrect and hesitant words receive priority in later practice. A minimal learning-data workflow supports review of these signals.
Results became more diagnostic
The result screen goes beyond a simple score. During the question phase, the semantic image stays hidden until the answer is confirmed, then remains visible until the next question. Pre-learning cards remain a separate part of the flow.
V2.0 / Interface evidence
A result that explains the round.
The results view brings accuracy, recognition classifications, average effective recognition time, and review words together.
This captured interface shows a ten-word automated verification round: 100% accuracy, 10 fast recognitions, and a 0.4-second average effective time. These values are synthetic test data, not a student performance record.
Archive / V1.1 interface
Earlier screens document the product’s previous stage; they are not presented as the current V2.0 interface.
02 / Product concept
A short loop, designed to be repeated.
Letter World turns recognition practice into a short, repeatable loop: encounter a prompt, make a choice, receive immediate feedback, and return to difficult material. The product is designed to make repetition feel purposeful without hiding the learning task behind a large game world.
- 01Select a level
- 02Choose 10–30 words
- 03Answer under time pressure
- 04Review feedback
- 05Repeat difficult material
03 / My role
Founder and Product Lead
Founder and Product Lead: product direction, learning goals, implementation coordination, version testing, and feedback-driven iteration.
This role describes project initiation, product direction, coordination, testing, and iteration. It does not imply that I wrote every part of the engineering system alone.
04 / Development and iteration
Make the learning loop clearer, not the world larger.
“I gradually learned that a useful educational product did not need a larger virtual world. It needed a clearer and more repeatable learning loop.”
V1.1 brought the game client, question bank, content-management workflow, and a validated release artifact together. The archived V1.1 artifact is an earlier milestone. V2.0 is the current live release, as confirmed by the project owner; this website review did not independently probe the production service.
V1.1 / An integrated release
Established the client, content-management workflow, and repeatable learning loop.
V2.0 / Recognition and diagnosis
Added selectable rounds, structured generation, recognition timing, review priorities, and anonymous aggregate learning-data administration.
05 / Student testing
Repeated use, with a clear evidence boundary.
Ten students tried V2.0 on launch night and gave positive feedback. Across five consecutive school days, 18 classmates used it daily, each completing at least two rounds per day. Among those users, 10 specifically said that ten-word rounds suited short breaks, felt quick, and helped them notice words more readily. These are their reported experiences, not a measured learning gain.
These observations document real product use and qualitative feedback. They do not constitute a controlled study or demonstrate a measured improvement in academic performance.
Continuing vocabulary priorities from earlier feedback
Raise the difficulty of the high-school vocabulary bank.
Reduce overly simple foundational words in later selections.
Add high-frequency error-prone words and words with multiple meanings.
Include easily confused word pairs and difficult words common in reading.
06 / Technical system
One learning experience, four connected surfaces.
The product connects a mobile learning experience with content management, application data, and a validation pipeline.
Cocos Creator + TypeScript
Mobile H5 game flow, questions, audio, images, and immediate feedback.
Vue + Vite admin
Question-bank review, editing, state changes, and publishing controls.
ThinkPHP + MySQL
Multi-application API and published-section delivery to the learning client.
Python validation flow
Draft generation, JSON-schema checks, quality validation, and local preview before import.
Verified stack: Cocos Creator 2.4.11, TypeScript, Vue 3, Vite, ThinkPHP 8, PHP 8+, MySQL, and a Python 3.10+ content pipeline.
07 / Reflection
Product quality depends on what gets removed, too.
Building Letter World changed how I think about educational technology. A feature can be technically complete and still create friction for a learner. Version testing made interface clarity, vocabulary quality, and repetition design as important as adding new systems.