AI/ML工程師 | AI/ML Engineer
Description
【職位概述】
Tricuss (十論科技) 致力於顛覆傳統研發流程(Revolutionize R&D)。旗艦產品「企業級 Co-researcher AI Agent」為自主虛擬科學家,融合 Agentic AI、大數據與物理模擬,能自主設計假說、推演,並自動檢索全球論文與內部資料以加速研發探索。
我們將系統與「物理基礎數位孿生(Physics-based Digital Twins)」深度結合,專為地端(On-premise)與封閉網路(Air-gapped)高資安客戶打造,實現最高千倍加速並減少 70% 實體實驗。目前我們與 NVIDIA、Qualcomm、Intel 及世界頂尖半導體與 AI 伺服器供應鏈的客戶們密切合作,邀您共同定義下一代研發標準。
身為 AI 工程師,您將是打造我們旗艦產品「企業級 Co-researcher AI Agent」的核心推手。我們的產品不僅僅是回答問題的聊天機器人,而是具備深度邏輯推理、自主實驗與數據探勘能力的虛擬資料科學家與研究夥伴。您的挑戰在於研發並優化結合大型語言模型 (LLM) 與傳統機器學習 (ML) 的複合型 AI 代理 (AI Agent),使其能自主提出假設、設計實驗、操作 ML 模型,並與企業內部由 Hadoop、Apache Iceberg、Apache Spark 及 ETL 構建的大數據生態系進行深度交互。您必須確保這個強大的系統能在嚴格的合規標準與封閉網路下安全運行,為客戶產出具備極高商業與研發價值的洞察。
【主要職責】
- Co-researcher AI 代理大腦開發: 設計並實作具備複雜研發任務拆解、長期記憶與跨領域邏輯推理能力的 Co-researcher 核心架構。確保 AI 代理能自主理解企業業務痛點,主動制定研究計畫、調用各類分析工具,並綜合歸納出具策略意義的洞察報告。
- ML Agent 研發與自動化機器學習 (AutoML) 整合: 開發專注於機器學習任務的 ML Agent,使其能自主執行資料預處理、特徵工程 (Feature Engineering)、模型訓練、超參數調優與模型評估。將傳統 ML 框架(如 Scikit-learn, XGBoost, PyTorch)無縫封裝為 AI 代理可靈活調用的工具 (Tools/Skills)。
- 大數據生態系與 AI 管線深度結合: 與資料工程及後端團隊緊密合作,利用 ETL 與 Apache Spark 等工具建立自動化的資料萃取與清洗管線。確保 Co-researcher 與 ML Agent 能即時存取、分析並處理儲存於 Hadoop 與 Iceberg 中的海量結構化與非結構化企業級數據。
- 進階 AI 檢索系統與向量工程: 針對企業內部專有文獻、研究報告與海量日誌,設計並優化檢索系統。建置高效能的向量資料庫 (如 Milvus, Qdrant),實作混合檢索、語意路由與重排序策略,為 Co-researcher 提供精準的知識檢索底層支援。
- 地端大語言模型 (LLM) 部署與推論優化: 針對無法連網的高資安地端環境,負責開源大語言模型的評估與私有化部署。利用 vLLM 等推論引擎進行模型量化與效能調優,確保 Co-researcher 在邊緣或地端伺服器上能維持極低的延遲與高吞吐量。
- AI 與 ML 資訊安全防護實作: 將資安思維融入 AI 與 ML 模型開發的每個環節。實作企業級資料脫敏、個人機密資訊 (PII) 過濾機制,防範模型毒化 (Data Poisoning) 與提示詞注入 (Prompt Injection),確保 AI 代理的所有研究行為與資料存取絕對符合企業權限規範 (RBAC) 與零信任架構。
【Position Overview】
As an AI Engineer, you'll drive this flagship product—not a chatbot, but a virtual data scientist with deep reasoning, autonomous experimentation, and data-mining skills. You'll build and optimize a hybrid AI Agent combining LLMs with traditional ML that autonomously proposes hypotheses, designs experiments, operates ML models, and interacts deeply with the enterprise big-data ecosystem (Hadoop, Apache Iceberg, Apache Spark, ETL)—running securely under strict compliance and air-gapped networks to produce high-value commercial and R&D insights.
【Key Responsibilities】
- Co-researcher Agent "Brain" Development: Build the core architecture for decomposing complex R&D tasks, long-term memory, and cross-domain reasoning, so the Agent can grasp enterprise pain points, formulate research plans, invoke analytical tools, and synthesize strategic insight reports.
- ML Agent and AutoML Integration: Build an ML Agent that autonomously runs data preprocessing, feature engineering, model training, hyperparameter tuning, and evaluation, wrapping traditional ML frameworks (Scikit-learn, XGBoost, PyTorch) into tools/skills it can flexibly invoke.
- Big-Data Ecosystem and AI Pipeline Integration: Partner with data engineering and backend teams to build automated extraction and cleansing pipelines (ETL, Apache Spark), enabling the Co-researcher and ML Agent to access and process massive structured and unstructured data in Hadoop and Iceberg in real time.
- Advanced AI Retrieval and Vector Engineering: Design and optimize retrieval over proprietary literature, reports, and logs; build high-performance vector databases (Milvus, Qdrant) with hybrid retrieval, semantic routing, and re-ranking for precise knowledge retrieval.
- On-premise LLM Deployment and Inference Optimization: Evaluate and privately deploy open-source LLMs for offline, high-security environments, using engines such as vLLM for quantization and tuning to keep latency low and throughput high on edge or on-premise servers.
- AI and ML Security: Embed security into every stage of development—enterprise-grade data masking and PII filtering, defense against data poisoning and prompt injection—ensuring all Agent research and data access comply with enterprise permissions (RBAC) and zero-trust architecture.