AI 工程師 / AI Engineer

本職位負責跨產品線的 AI 系統開發與整合，核心工作涵蓋三大領域：影像辨識、機器學習預測系統、以及 LLM/Agent 應用開發。不侷限於單一技術領域，需具備將各類 AI 技術落地到實際產品與內部流程中的能力。

This role is responsible for cross-product-line AI system development and integration, spanning three core domains: computer vision, machine learning prediction systems, and LLM/agent application development. Rather than specializing in a single domain, this engineer will apply a broad range of AI techniques into both production systems and internal workflows.

【主要工作內容 / Key Responsibilities】

1. YOLO 視覺辨識模型訓練與優化 / Vision Detection Model Training & Optimization
・基於 Ultralytics YOLO 進行魚群 / 船隻目標偵測模型的訓練、評估、超參數調優
Train, evaluate, and tune hyperparameters for fish/vessel detection models using Ultralytics YOLO

・模型推論優化與匯出（ONNX / TensorRT），支援邊緣裝置部署
Optimize inference performance and export models (ONNX / TensorRT) for edge deployment

・管理 CVAT 標注後處理流程與訓練資料品質
Manage CVAT annotation post-processing workflows and training data quality

2. 機器學習漁場預測模型訓練與優化 / Machine Learning Fishing Ground Prediction
・整合衛星環境資料（SST、葉綠素、洋流）與船載感測器訊號（聲納、雷達），設計 ETL 與特徵工程管線
Integrate satellite environmental data (SST, chlorophyll, currents) with onboard sensor signals (sonar, radar); design ETL and feature engineering pipelines

・訓練與優化 CPUE / 漁場預測模型，結合時序、空間與環境變數
Train and optimize CPUE / fishing ground prediction models incorporating temporal, spatial, and environmental variables

・開發邊緣端推論模組，支援每日漁場建議與離線容錯機制
Develop edge inference modules for daily fishing ground recommendations and offline fallback mechanisms

3. LLM 應用開發 / LLM Application Development
・使用 Google ADK (Agent Development Kit)、Gemini SDK 開發 Agent 應用，串接內部系統與工作流程
Build agent applications using Google ADK (Agent Development Kit) and the Gemini SDK, integrating with internal systems and workflows

・使用 n8n 設計自動化流程，串接 LLM 與既有系統（如文件生成、資料處理、通知系統）
Design automation workflows using n8n to connect LLM capabilities with existing systems (e.g., document generation, data processing, notifications)

・評估並導入 RAG、Prompt Engineering 等技術於內部效率工具與客戶端功能
Evaluate and implement RAG, prompt engineering, and related techniques for internal efficiency tools and customer-facing features