Senior Associate/Assistant Vice President, Full Stack Agentic AI Engineer

Introduction

Agentic AI engineering is a frontier discipline that requires both deep AI system knowledge and the full-stack engineering rigour to ship production-grade systems that operate reliably at enterprise scale. Building an agent that works in a notebook is table stakes; building one that handles edge cases gracefully, integrates securely with enterprise systems, and can be monitored and maintained by a team is the real engineering challenge. 

The Full Stack Agentic AI Engineer at Temasek is a hands-on builder responsible for designing, building, and operating AI-powered products and agent workflows across Temasek's investment and operations functions. You will work end-to-end — from LLM integration and agent orchestration through to front-end interfaces, API integrations, and production monitoring — with a relentless focus on reliability, observability, and user experience. You will be building systems that investment professionals and operations teams depend on for real decisions, not demos.

Responsibilities

Agentic AI system engineering

• Design and implement multi-step agentic AI workflows: agent orchestration (single-agent and multi-agent coordination), tool-calling architectures, memory and state management, and planning and reasoning loops using frameworks such as LangGraph, AutoGen, or custom orchestration.
• Build production-grade RAG (Retrieval-Augmented Generation) pipelines: document ingestion and chunking strategies, embedding generation and management, vector store integration (Pinecone, Weaviate, pgvector, or equivalent), semantic search, and hybrid retrieval architectures.
• Implement robust tool and API integration layers that allow AI agents to safely interact with enterprise systems — including investment data platforms, portfolio management systems, market data APIs, and internal knowledge bases — with appropriate access controls, rate limiting, and error handling.
• Engineer agent prompting and context management systems: system prompt architecture, dynamic context construction, prompt versioning, and evaluation-driven prompt optimisation using automated and human evaluation frameworks.
• Contribute to the design and extension of Temasek's MCP (Model Context Protocol) server ecosystem — building new tool connectors and extending the platform's tool library with reusable, well-documented integrations.

Full-stack product engineering

• Build the full application stack for AI products: React or equivalent front-end interfaces designed for professional user workflows; Python or Node.js back-end services; RESTful and streaming API layers; and enterprise integration middleware.
• Design and implement real-time AI interaction patterns: streaming LLM responses, progressive disclosure of agent reasoning steps, asynchronous long-running agent task management, and WebSocket-based live update interfaces.
• Own API design and documentation for AI product services — ensuring clean contracts, versioning discipline, and integration documentation that enables downstream consumers (other internal systems, citizen developer tools) to build reliably on your services.
• Collaborate with data engineers to design the data access layer for AI products: query optimisation for AI-readable data, structured data extraction from unstructured sources, and data pipeline integrations that keep AI product knowledge bases current.

Production engineering and reliability

• Instrument all AI systems with comprehensive observability: structured logging of agent actions, LLM call tracing (latency, token usage, model version), error rate monitoring, and alerting calibrated to the failure modes specific to agentic systems.
• Design for graceful degradation: ensure AI systems fail safely, fall back to human escalation paths when uncertain, and communicate their limitations clearly to users — critical in a financial services context where AI failures have real consequences.
• Implement testing strategies appropriate for non-deterministic AI systems: unit tests for deterministic components, evaluation suites for LLM behaviour (using frameworks such as LangSmith Evals, RAGAS, or custom evaluation pipelines), regression testing for prompt changes, and integration testing across the full agent workflow.
• Participate actively in code review, architecture discussions, and production incident response — with a culture of collective ownership over system reliability and continuous improvement.

Requirements

Experience and background

• 4–8 years of software engineering experience with strong full-stack depth and at least 2 years focused on AI/ML system engineering — ideally building and operating agentic AI or LLM-powered products in production.
• Hands-on experience with modern agentic AI frameworks: LangChain, LangGraph, AutoGen, CrewAI, or equivalent; strong familiarity with the Anthropic and OpenAI SDKs and their tool-calling and streaming APIs.
• Demonstrated experience shipping production AI systems with real users — not research experiments, but deployed products with monitoring, incident response, and continuous improvement cycles.
• Experience in financial services, fintech, or data-intensive regulated industries is advantageous.

Technical capabilities

• Languages and frameworks: Python (primary, including FastAPI, asyncio, Pydantic), TypeScript/JavaScript (React, Next.js, Node.js); comfortable across the stack without deep specialisation limiting your ability to contribute end-to-end.
• AI/ML stack: LLM integration (Anthropic Claude, OpenAI GPT-4o, Google Gemini), embedding models, vector databases, RAG pipeline components, evaluation frameworks (LangSmith, RAGAS, custom), and observability tooling (Helicone, LangFuse, or equivalent).
• Infrastructure and DevOps: containerisation (Docker, Kubernetes), cloud deployment (AWS, Azure, or GCP), CI/CD pipelines, secrets management, and basic security hygiene for AI systems (input sanitisation, output validation, access-scoped tool permissions).
• Data engineering basics: SQL proficiency, familiarity with data pipeline tooling (dbt, Airflow, or equivalent), and ability to design data access patterns optimised for AI consumption.