WEBSPARKS PTE. LTD.
ML/AI Engineer - Fraud Analytics Model (Public Sector)
- Compensation
- SGD 8K–SGD 12K / mo
- Location
- Singapore
- Work setup
- Hybrid
- Seniority
- Not specified
- Employment
- Contract
- Experience
- 3+ years
- Category
- Ml / AI
- Posted
Skills
Nice to have
- Experience with AWS services such as Bedrock, Redshift, Glue, S3, Lambda or related cloud-native services.
- Experience with PySpark, Databricks, large-scale data processing or feature engineering pipelines.
- Experience with CI/CD, containerisation, MLOps or model deployment tools such as Docker, MLflow, GitHub Actions or Jenkins.
- Experience implementing monitoring, observability and operational support processes for ML/AI applications.
- Experience delivering fraud-detection, risk, compliance, investigation or decision-support solutions.
- Experience with knowledge retrieval, retrieval-augmented generation (RAG), graph-based workflows, prompt management, model evaluation tools or AI observability platforms.
- Familiarity with Singapore Government or public-sector technology environments, including data governance, security and compliance requirements.
- Experience working with cross-functional teams of data scientists, engineers, analysts and business stakeholders.
Full description
- 1-year contract, renewable
- Government project
- Hybrid work arrangement
Role Overview
The ML/AI Engineer will join the Fraud Analytics Model (FAM) team within the Data Analytics and AI Division (DSAD). The role supports FAM 2.0, which detects misuse of public training subsidies, by designing and building two agentic AI systems.
One system will surface emerging fraud risks and blind spots. The other will help turn validated risks into actionable detection logic. Human validation, evaluation and responsible AI guardrails will be built into both systems.
You will deliver working, evaluated minimum viable products (MVPs) and help build DSAD’s in-house capability to operate and maintain them independently.
Key Responsibilities
- Agentic AI architecture: Design the end-to-end architecture for FAM agentic AI solutions, including agent orchestration, workflows, state management, tool integration, human approval points and system interactions. Use suitable frameworks such as LangGraph or CrewAI. Architecture decisions will be made in consultation with DSAD’s Platform Team.
- Emerging-risk agent development: Develop AI agents that analyse approved information sources to identify, generate and prioritise evidence-linked emerging fraud and abuse risks for officer review.
- Detection engineering agent development: Develop AI agents that assess data feasibility, propose candidate risk indicators and detection features, and support the translation of validated risks into actionable analytics and detection logic.
- Data and analytics integration: Integrate agent workflows with FAM 2.0 data and analytics components, including Redshift, Glue and PySpark. Build and maintain the data preparation and feature engineering processes needed to support AI and ML workloads.
- Production engineering: Turn prototypes into modular, tested, maintainable and deployable services, workflows, APIs or agentic applications, with clear interfaces, version control and operational ownership.
- CI/CD and automation: Build automated workflows to test, validate and promote code, prompts, workflows and AI components across environments. Include quality, security and human approval gates where required.
- Evaluation and AI guardrails: Implement evaluation frameworks and governance controls covering output quality, reproducibility, traceability, access control, exception handling, human approval and safeguards against untrusted inputs, such as prompt injection.
- Monitoring and reliability: Implement observability, dashboards and alerts for system health, latency, throughput, errors, resource use, data quality, workflow performance and operating costs. Support incident management, troubleshooting and root-cause analysis.