Hanzi Flashcards: Triple-Sided Chinese Learning App

The Challenge: The “Flashcard Creation” Bottleneck

Learning Chinese is hard. Creating flashcards is harder. Traditional apps force users to manually type characters, look up pinyin, and find the right translation. For a student reading a 500-word article, this means 3 hours of manual card creation.

The goal was: Paste a news article and get a custom, high-quality study set in under 10 seconds.

The RAG-Powered Lexicon

We didn’t just ask an LLM to “translate words” (which leads to hallucinations in niche contexts). We built a RAG-based parsing engine.

1. The HSK/CEDICT Vector Store

We ingested the entire HSK (Hanyu Shuiping Kaoshi) curriculum and the CEDICT open-source dictionary into a structured Vector Database.

2. Contextual Tokenization

When a user pastes text, our system doesn’t just look for words—it analyzes the grammar and retrieves the most “contextually appropriately” definitions from the RAG store. This ensures that a character like 行 is correctly identified as “walk” (xíng) or “row” (háng) based on the surrounding text.

3. Triple-Sided Generation

The AI automatically populates the specialized Hanzi-Pinyin-English triangle, ensuring every card is perfect for recognition, production, and pronunciation practice.

Technical Architecture

Frontend: Astro + React PWA (Native-like experience on mobile).
Backend: Node.js API with LangChain orchestration.
RAG Store: SQLite-VSS for lightweight, lightning-fast dictionary lookups.
Deployment: hanzi-flashcards.com and the Apple App Store.

Business ROI

Productivity: Reduced card creation time by 95% (from 3 minutes per card to <1 second).
Retention: Users create 5x more cards than in traditional apps, leading to higher long-term engagement.
Model: Successfully launched as a Freemium app with Pro tiers for advanced RAG features.

Ready to build your own AI-powered tool? Check our AI Readiness Checklist.