Langquence API

A FastAPI-based English correction service designed for Korean speakers preparing for English interviews. This service utilizes Large Language Models to analyze and improve English expressions, making them more natural and professional.

Project Overview

Langquence API is part of a larger system designed to help Korean speakers improve their English interview skills. The system consists of multiple layers:

1. Data Collection Layer: Collects English interview data from sources like YouTube
2. Data Processing Layer: Processes the collected data including speech-to-text conversion
3. Model Learning Layer: Trains NLP models using the processed data
4. Realtime Processing Layer (this project): Provides real-time English correction via API
5. Presentation Layer: User interfaces for interacting with the service

This repository implements the Realtime Processing Layer as a FastAPI application.

Architecture

The application follows a modular architecture with clear separation of concerns:

langquence-was/
├── app/
│   ├── main.py              # FastAPI application entry point
│   ├── api/                 # API routes
│   ├── config/              # Configuration management
│   ├── consts/              
│   ├── clients/             # External service clients
│   ├── services/            # Business logic
│   ├── dto/                 
│   └── utils/               
├── .env                     # Environment variables
└── requirements.txt         # Dependencies

Key Components

• API Layer: Handles HTTP requests and responses
• Correction Service: Orchestrates the correction process
• Qwen Client: Interacts with Alibaba Cloud's Qwen LLM API
• Pattern Matching Engine: Validates and enhances LLM responses
• Feedback Generator: Creates user-friendly feedback from corrections

Getting Started

Prerequisites

• Python 3.8+
• An Alibaba Cloud account with access to DashScope API

Installation

1. Clone the repository:

git clone https://github.com/your-username/langquence-was.git
cd langquence-was

2. Create and activate a virtual environment:

python -m venv venv

# Windows
venv\Scripts\activate

# Mac/Linux
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Create a .env file in the project root with the following contents:

APP_NAME=Langquence API
APP_EXECUTE_COMMAND=app.main:app
APP_HOST=127.0.0.1
APP_PORT=8000
APP_RELOAD=True
API_PREFIX=/api
DEBUG=True
LOG_LEVEL=INFO
ALIBABA_API_KEY=your_api_key_here
MODEL_NAME=qwen-plus
MAX_TOKENS=512

Running the Application

Run the application using Uvicorn:

uvicorn app.main:app --reload

Or use the provided entry point:

python -m app.main

The API will be available at: http://localhost:8000

API Usage

Correction Endpoint

Corrects English text for interview settings.

Endpoint: POST /api/correct

Request:

{
  "text": "I have work in this company for 5 years."
}

Response:

{
  "original": "I have work in this company for 5 years.",
  "needs_correction": true,
  "corrected": "I have worked in this company for 5 years.",
  "explanation": "Changed 'work' to 'worked' to use present perfect tense correctly for an action that started in the past and continues to the present.",
  "alternatives": ["I have been working in this company for 5 years."]
}

Architecture Details

Pattern Matching Engine

The Pattern Matching Engine validates and enhances the LLM's corrections by:

1. Checking if the LLM identified errors correctly
2. Comparing against known error patterns
3. Providing additional validation for the corrections
4. Suggesting alternative expressions when appropriate

Prompting Strategy

The application uses a carefully designed prompting strategy that:

1. Provides clear instructions to the LLM
2. Specifies the expected response format
3. Includes examples of common errors and corrections
4. Guides the model to produce consistent, high-quality responses

Development

Adding New Features

To add new features to the application:

1. Implement traffic rate limitation
2. Enhance security policies
3. Optimize prompt token usage
4. Refactoring the Architecture
5. Connecting to the DB for saving requests

Testing

To run tests:

pytest