Senior AI Engineer

Tổng quan về công việc và trách nhiệm

We are scaling agentic AI across the enterprise using a multi-platform agent stack comprising Microsoft Copilot Studio, Google AgentSpace, AWS Bedrock, and Dataiku. This role designs, builds, and operates production-grade AI agents that can: - reason and plan, - retrieve enterprise knowledge, - take actions via tools and APIs, - operate safely under governance and audit constraints. Agents are expected to go beyond chat and support real business workflows across IT, operations, and other functions. The role is platform-aware but not platform-locked. Candidates may come from any ecosystem, but must demonstrate the ability to deliver agentic systems on at least one of the supported platforms and adapt across the rest. What you will build Examples of solutions you will deliver include: - Enterprise Support and Ops Agents • Leverage AI and machine learning to learn from past tickets, emails and logs to triage requests and monitor systems • Propose or execute actions with escalation and full audit trails • Perform automated root cause analysis for complex enterprise environment - Knowledge and Policy Agents • Grounded Q&A over SOPs, manuals, and policies • Citations, traceability, and access-controlled retrieval • Security attestation agent to reduce administrative work and to guide users for policy compliance • System onboarding agent to guide users and help in troubleshooting, price and cost estimation - Workflow and Action Agents • Multi-step orchestration (e.g. read request → analyze → fetch data approval → update system) Responsibilities 1. Agentic Solution Design - Design goal-driven agents that decompose tasks, select tools, manage state, and recover from failures. - Implement agent patterns such as planner–executor, coordinator–worker, reflection/self-check, and human-in-the-loop decision gates. 2. Platform Implementation You will work across one or more of the following platforms: - Microsoft Copilot Studio • Build copilots with plugins, connectors, and enterprise guardrails - Google AgentSpace • Build agents integrated with other applications or tools (e.g. ITSM, SIEM, Workflow management system) • Orchestrate multi-step workflows and API-driven actions - AWS Bedrock • Design secure agentic workflows using Bedrock models and tools - Dataiku • Operationalize agents within analytics, ML pipelines, and business workflows - An operational Hybrid Machine Learning and LLM model to support smart Digital Platform and Security Operations • ML models detect anomalies → LLM explains them • ML predicts incidents → LLM drafts remediation and healing steps • ML scores risk → LLM supports human decision-making 3. Retrieval and Grounding (RAG) - Design enterprise RAG pipelines including ingestion, chunking, embeddings, retrieval, reranking, and citation. - Ensure retrieval respects role-based access control and data classification. 4. Evaluation, Observability, and Operations - Build evaluation frameworks for non-deterministic systems: task success metrics, grounding checks, hallucination detection, and regression tests. - Implement observability for prompts, retrieval, and tool calls. - Own solutions from POC through MVP and production. 5. Security, Governance, and Responsible AI - Enforce least-privilege tool access, audit logging, secrets management, prompt-injection defenses, and safe action boundaries. - Design human approval checkpoints for high-risk or irreversible actions. - Comply with enterprise AI governance and ethics requirements.