Ryan Ruan

Key Skills

Software

Appen

CVAT

LabelImg

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Top Task Types

Classification

Polygon

Polyline

Text Generation

Translation Localization

Freelancer Overview

My main experience lies in the data annotation of various small models such as CV/OCR models, including YOLO and PaddleOCR models. I have my own understanding of the upper limits and advantages of the capabilities of various models. The trained models have been applied in the industrial field. In addition, I have also conducted annotation work for LLM models on the Appen platform, evaluating the generated content and performing local editing.

Labeling Experience

text2sql LLM

Internal Proprietary ToolingTextFine Tuning

Fine-tune the Qianw LLM model, integrate with the ERP API, and enable natural language to query SQL language data.

Fine-tune the Qianw LLM model, integrate with the ERP API, and enable natural language to query SQL language data.

2025 - 2025

Automatic detection of pump status

CVATImageBounding BoxPolygon

A water pump factory needs a software to automatically check whether their water pump products are correctly connected to the water pipes. The factory has provided 5,000 on-site photos of the water pumps and pipes. By combining multiple technologies, we ensure the accuracy. We trained the water pump and water pipe separately, and marked the connection points. The final model's mAP is approximately 0.6. We implemented the solution by integrating the multimodal large model API.

A water pump factory needs a software to automatically check whether their water pump products are correctly connected to the water pipes. The factory has provided 5,000 on-site photos of the water pumps and pipes. By combining multiple technologies, we ensure the accuracy. We trained the water pump and water pipe separately, and marked the connection points. The final model's mAP is approximately 0.6. We implemented the solution by integrating the multimodal large model API.

2025 - 2025

flame detection

Internal Proprietary ToolingImagePolygon

Combining the heat sensor with monitoring, mark the areas with flames, and achieve real-time automatic identification of flames by the monitoring system, thereby triggering fire alarm when a fire is detected.

Combining the heat sensor with monitoring, mark the areas with flames, and achieve real-time automatic identification of flames by the monitoring system, thereby triggering fire alarm when a fire is detected.

2025 - 2025

workplace regulation recognition

CVATImageBounding BoxPoint Key Point

The factory requires automatic detection robots to automatically monitor the discipline conditions in the production workshop. 1: Whether the inventory has been placed in the correct location. Finally 2: Whether there are any debris piled up in specific areas. 3: Whether there is any uncleaned garbage or pollution. 4: Others. The total data of each scenario is about 3,000 images, and the final model's mAP is approximately 0.8.

The factory requires automatic detection robots to automatically monitor the discipline conditions in the production workshop. 1: Whether the inventory has been placed in the correct location. Finally 2: Whether there are any debris piled up in specific areas. 3: Whether there is any uncleaned garbage or pollution. 4: Others. The total data of each scenario is about 3,000 images, and the final model's mAP is approximately 0.8.

2025 - 2025

LLM generates scores/localization

AppenTextPrompt Response Writing SFT

Score the content generated on the LLM, and convert the generated English content into Chinese content.

Score the content generated on the LLM, and convert the generated English content into Chinese content.

2024 - 2025

Education

Work History