LexiMind: An AI TOEFL English Learning System Based on Structured Instructions

LexiMind is a web application designed specifically for TOEFL test-takers. Through structured interactions with large language models (LLMs), it helps learners improve vocabulary retention, phrase usage, and writing expression. Its core design philosophy is instruction-driven, requiring users to input content in predefined formats. This not only ensures output consistency but also significantly reduces token consumption, enabling cost-effective and efficient learning.

In the era of artificial intelligence, traditional vocabulary books and grammar textbooks are gradually being replaced by intelligent tools. LexiMind is an innovative product in this trend. Targeting TOEFL preparation scenarios, it addresses the issues of traditional tools lacking personalization and efficient interaction. By combining structured instructions with LLMs, it provides more precise learning support.

LexiMind takes instruction-first as its core principle. All inputs must follow predefined command patterns; inputs that do not conform to the format will be rejected to avoid unnecessary API calls. Its instruction system covers the main needs of TOEFL preparation:

• Vocabulary learning: $ (English-only explanation), $cn (Chinese-English explanation), $cmp (word differentiation)
• Phrase learning: $$ (English-only phrase explanation), $$cn (Chinese-English phrase explanation)
• Writing improvement: $$$ (writing polishing and revision)
• Reading training: daily-reading (daily reading generation)
• General query: > (free interaction)

LexiMind uses a layered architecture:

• Frontend: Pure static HTML + JavaScript, high performance and low resource consumption;
• Backend: Python Flask/FastAPI, responsible for instruction parsing, LLM API calls, response formatting, etc.;
• Data layer: SQLite stores query history, learning trajectories, etc.;
• AI layer: Supports DeepSeek-V3 (SiliconCloud API) and Gemini 1.5 Flash. Security mechanisms include API key storage in environment variables, IP rate limiting, input length restrictions, etc.

LexiMind supports minimalist deployment, requiring only Python 3.10+ and a pip environment—no Docker or Nginx needed. The startup scripts (run.bat for Windows, ./run.sh for macOS/Linux) automatically create a virtual environment, install dependencies, configure API keys, start the local service (default: http://127.0.0.1:5000), and open the browser.

LexiMind is suitable for all stages of TOEFL preparation:

• Vocabulary accumulation: Build a vocabulary library via $/$cn, differentiate synonyms with $cmp;
• Phrase reinforcement: Master high-frequency phrases with $$/$$cn;
• Writing training: Get instant writing feedback with $$$;
• Daily reading: Maintain input with daily-reading;
• Pre-exam sprint: Use structured queries to quickly review key points.

Innovation: Encapsulates AI capabilities into a structured learning tool instead of a chatbot, enhancing learning purposefulness and reducing usage costs; Limitations: The strict instruction mode lacks flexibility for open discussions or complex interaction scenarios. However, this design aims to help learners stay focused—it is an intentional "feature" rather than a "bug".

LexiMind focuses on the core needs of TOEFL preparation. Through sophisticated instruction design, it achieves a balance between efficiency and cost, representing a pragmatic approach to AI education applications. As a lightweight AI learning assistant, it provides efficient and cost-effective learning support for TOEFL test-takers and is worth trying.