About
Hi, I’m Ramadan Ma 👋
Computer Application Engineer，专注于 AI 视觉应用与智能体系统的全链路开发——从数据标注、模型训练到生产部署。
能力方向
视觉 AI 应用
- YOLO 模型训练与微调 — 目标检测、姿态估计（Pose）、旋转框检测（OBB）
- 数据标注与数据集构建 — CVAT 帧级标注、负样本增强、数据集管理
- 推理系统开发 — 实时视频流推理、多模型协同、结果异步上报
智能体 & LLM 应用
- RAG 系统开发 — 私有化知识库部署、向量检索调优、权限隔离设计
- 智能体节点开发 — Function Call、MCP、Agent、HTTP Request、SQL 等核心节点
- 工作流平台集成 — Dify / N8N / Langflow 技术选型与系统集成
项目经历
员工消毒合规实时检测系统
基于骨骼关键点（踝点坐标）与消毒区 ROI 的时序状态机，实时判断员工进入场区时是否完成足够时长的消毒（≥1 秒）。结合人脸识别模块完成身份绑定，检测结果异步推送管理后台供复核。针对门框易被误识别为人腿的问题，基于 Objects365 数据集对 YOLOv8-Pose 进行微调，检测精度达 90%+。
企业内部私有知识库问答系统
基于 RAGFlow 开源框架进行私有化二次开发，搭建企业内部 LLM 问答系统。设计用户与文件权限隔离的数据库表结构，使用 Python 开发用户管理与文件管理接口，对 100+ 文档知识库完成系统性问答测试与向量检索效果调优，并对外维护对话 API 供公司内部其他系统集成调用。
员工操作规范合规检测系统
使用 CVAT 对现场监控视频进行帧级数据标注，选型并训练 YOLOv8-OBB（旋转框检测）模型。相比普通矩形框，OBB 能更精准适配人体姿态的方向变化，有效降低误检率。独立完成从数据标注、模型训练到推理脚本部署的全流程。
企业智能体平台技术选型与节点开发
系统调研并本地部署 Dify、N8N、Langflow 三款主流开源智能体/工作流平台，完成横向对比并输出技术选型报告。主导节点体系结构设计，统一定义各类节点的输入输出规范；独立开发全部节点类型（Function Call、MCP、Agent、HTTP Request、SQL），辅助 C# 后端团队完成公司内部智能体平台集成方案。
已发布应用
RanTerm — SSH & SFTP Client for macOS
独立开发并上架 Mac App Store 的 macOS 终端工具，支持 SSH 连接与 SFTP 文件管理，原生 macOS 体验。
技术栈
联系我
Hi, I’m Ramadan Ma 👋
Computer Application Engineer focused on the full lifecycle of AI vision applications and intelligent agent systems — from data annotation and model training to production deployment.
What I Do
Visual AI Applications
- YOLO Model Training & Fine-tuning — Object detection, pose estimation, oriented bounding box detection (OBB)
- Data Annotation & Dataset Construction — Frame-level annotation with CVAT, negative sample augmentation, dataset management
- Inference System Development — Real-time video stream inference, multi-model coordination, async result reporting
AI Agent & LLM Applications
- RAG System Development — Private knowledge base deployment, vector retrieval tuning, user-file permission isolation
- Agent Node Development — Function Call, MCP, Agent, HTTP Request, SQL and other core node types
- Workflow Platform Integration — Tech selection and system integration with Dify / N8N / Langflow
Project Experience
Employee Disinfection Compliance Detection System
Built a real-time compliance detection system using a skeletal keypoint (ankle coordinate) and disinfection zone ROI time-series state machine, determining whether employees complete sufficient disinfection time (≥1 second) when entering a facility. Integrated face recognition for identity binding; detection results are pushed asynchronously to a management backend for review. Fine-tuned YOLOv8-Pose on the Objects365 dataset to address door-frame misidentification, achieving 90%+ accuracy.
Enterprise Private Knowledge Base Q&A System
Privately deployed and customized RAGFlow to build an internal enterprise LLM Q&A system. Designed a database schema for user-file permission isolation, developed user and file management APIs in Python, conducted systematic Q&A testing on a 100+ document knowledge base with retrieval tuning, and maintained the conversation API for integration by other internal business systems.
Employee Operational Compliance Detection System
Performed frame-level data annotation on surveillance video using CVAT. Selected and trained a YOLOv8-OBB (oriented bounding box) model, which better adapts to human body pose direction changes than standard rectangular box detection, effectively reducing false positives. Completed the full pipeline — data annotation, model training, and inference deployment — independently.
Enterprise Agent Platform Tech Selection & Node Development
Systematically evaluated and locally deployed three mainstream open-source agent/workflow platforms — Dify, N8N, and Langflow — producing a tech selection report comparing workflow orchestration, API integration, model access, and private deployment. Led the node system architecture design, unified input/output specifications, and independently developed all node types: Function Call, MCP, Agent, HTTP Request, and SQL. Assisted the C# backend team in completing the company’s internal agent platform integration.
Published Application
RanTerm — SSH & SFTP Client for macOS
An independently developed macOS terminal application supporting SSH connections and SFTP file management, available on the Mac App Store.
Tech Stack