New opportunity

Applied AI Engineer

Kuok (Singapore) Limited · Singapore

About this role

The Applied AI Engineer supports POSH’s AI programme by delivering prioritised use cases and developing AI agents and solutions on the Group’s enterprise AI platform. The role spans business engagement, prototyping, deployment and user adoption; prior domain knowledge is not required.

Responsibilities

Work with operational teams to understand workflows, identify AI opportunities and translate business needs into technical specifications and prototypes. Facilitate user acceptance testing, capture feedback, manage expectations and hand solutions over with documentation and user guides.
Assess data quality and schemas, identify dependencies and integrate solutions with enterprise systems. Develop human-in-the-loop review interfaces with domain experts.
Configure AI agents using patterns such as multi-agent architectures, tool use and autonomous workflows. Build RAG pipelines covering ingestion, chunking, embeddings, vector indexing and retrieval optimisation; integrate LLMs using prompts and structured outputs. Develop document-processing components using OCR and multimodal extraction.
Help evaluate accuracy, hallucinations and retrieval quality; test agents for edge cases, adversarial inputs and failure modes before deployment. Monitor deployed solutions, escalate anomalies and platform gaps, and document configurations, schemas and integrations to enterprise standards.

Requirements

1–2 years of software engineering or data/AI engineering experience, including exposure to LLM development. Internships, research attachments and early-career roles count; final-year graduates with strong AI portfolios are encouraged to apply.
Hands-on experience taking at least one AI or ML project from concept to working output; production deployment is advantageous, not required. Experience or demonstrated interest in working with non-technical users is expected.
Proficiency in Python; working knowledge of RAG, prompt engineering and structured outputs; familiarity with agentic patterns and orchestration; exposure to cloud-native development and vector databases; understanding of OCR, PDF/image extraction, chunking, multimodal content and API and microservices integration. Microsoft Azure is preferred.

Advantageous: Knowledge of maritime, offshore or comparable regulated, asset-heavy industries; compliance environments; and SharePoint, Microsoft 365, SAP or equivalent enterprise platforms. The role requires clear communication, structured thinking, ownership, comfort with evolving requirements, and rigorous documentation and testing.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

PythonAI agentsMulti-agent architecturesTool useHuman-in-the-loop workflowsRAGPrompt engineeringStructured outputsLLM integrationVector databasesMicrosoft AzureOCRPDF extractionImage extractionMultimodal content processingREST APIsAsynchronous APIsMicroservicesData integrationUser acceptance testingAI evaluationHallucination detectionRegression testingSharePointMicrosoft 365SAPStakeholder communicationRequirements gathering

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Turn operational needs into usable solutions
The role works with POSH teams to understand workflows, validate AI opportunities, build prototypes, run user acceptance testing, and support handover. The posting gives business engagement and user adoption equal weight with technical execution.
Make data and integrations workable
Responsibilities include checking data quality and schemas, identifying dependencies early, and connecting solutions to enterprise systems using established integration standards.
Build agents and knowledge workflows
The role configures agents on the Group’s enterprise AI platform and develops RAG pipelines, LLM integrations with structured outputs, and OCR or multimodal document-processing components.
Test, monitor and document AI behaviour
The role helps evaluate accuracy, hallucinations and retrieval quality; tests agents for edge cases and failures before deployment; monitors deployed solutions; and documents configurations and integrations to enterprise standards.

Likely challenges to think through

Balancing a useful prototype with real workflow needs
Preparation hypothesis, not a confirmed employer problem: evolving requirements could make a quick prototype diverge from what operational users need. Prepare to describe how you would validate the workflow, record assumptions, use UAT feedback and manage scope expectations.
Choosing where people should review agent outputs
Preparation hypothesis: more autonomous tool use may speed a workflow but increase the impact of incorrect outputs. Prepare to explain how you would choose human-in-the-loop checkpoints with domain experts and test edge cases before deployment.
Handling imperfect source data without hiding risk
Preparation hypothesis: retrieval and document extraction may be limited by data quality, schemas or system dependencies. Prepare to explain how you would surface those limits early, assess retrieval and extraction quality, and document unresolved dependencies rather than treating a working demo as deployment-ready.

Practice your answers

Technical practice

How would you assess whether operational data is ready for an AI use case?

What to coverPreparation suggestion: Explain how you would inspect data quality and validate schemas with data owners before building a Python pipeline. Describe how you would record assumptions and surface missing data or dependencies early.

How would you integrate an AI solution with an enterprise system?

What to coverPreparation suggestion: Discuss API contracts, REST or asynchronous interactions, error handling, and the organisation's integration standards. Explain how you would document schemas and integration points without assuming access to a particular enterprise platform.

When would you use tool use, multiple agents, or an autonomous workflow?

What to coverPreparation suggestion: Compare the patterns by task complexity, control, and failure risk rather than treating greater autonomy as inherently better. Explain how you would test agent behaviour before deployment.

How would you design a human-in-the-loop review step?

What to coverPreparation suggestion: Identify which outputs need domain-expert review and what context a reviewer needs to make a decision. Describe how feedback could be captured for iteration while keeping the review workable for operational users.

How would you build and improve a RAG pipeline for an operational knowledge base?

What to coverPreparation suggestion: Walk through ingestion, chunking, embeddings, vector indexing, and retrieval. Discuss how you would diagnose poor retrieval and compare changes using relevant test queries.

How would you make LLM outputs usable by a downstream application?

What to coverPreparation suggestion: Cover prompt design, a defined output structure, validation, and handling malformed or uncertain responses. Explain how you would refine the approach against operational and regulatory quality expectations.

How would you process PDFs or images for use in an AI solution?

What to coverPreparation suggestion: Describe OCR, extraction, preprocessing, and chunking, including how multimodal content might affect the result. Explain how you would detect extraction errors before passing content to retrieval or an LLM.

How would you evaluate an AI solution across model updates?

What to coverPreparation suggestion: Distinguish accuracy benchmarking, hallucination checks, retrieval-quality scoring, and regression tests. Explain how a representative test set and documented results would help reveal whether an update improved the use case.

How would you test an agent's failure modes before deployment?

What to coverPreparation suggestion: Describe structured tests for edge cases, adversarial inputs, and unexpected tool or workflow behaviour. Explain what evidence would prompt a fix, a human review step, or escalation.

What would you do when a deployed AI solution behaves unexpectedly?

What to coverPreparation suggestion: Explain how you would monitor for anomalies, capture enough context to investigate, and escalate promptly. Include how configuration and integration documentation would support resolution and handover.

Behavioural practice

Tell me about a time you learned a user's workflow before proposing a solution.

What to coverPreparation suggestion: Use a real situation to explain how you listened, clarified the problem, and translated it into a workable requirement or prototype. State what you did and what the user feedback or result showed.

Describe a time user feedback changed your work.

What to coverPreparation suggestion: Give the situation, how you presented or tested the work, and how you captured gaps without being defensive. Explain the change you made and its result; a structured test or review session is especially relevant.

Tell me about a time a dependency or blocker affected delivery.

What to coverPreparation suggestion: Explain the situation, how you identified the data, system, or timeline issue, and whom you informed or involved. Describe how you managed expectations and what happened next.

Describe a time you made progress while requirements were changing.

What to coverPreparation suggestion: Outline the competing priorities or uncertainty, the assumptions you documented, and the decision you made or escalated. Give the result and what you learned about balancing progress with validation.

Tell me about a time you handed work over so others could use or maintain it.

What to coverPreparation suggestion: Describe the situation, the documentation or walkthrough you provided, and how you checked that it met the users' needs. Explain the outcome without claiming production deployment if the work did not reach that stage.

Sharpen your knowledge

Topics to prepare

Operational use cases and user adoption
Posting: Business engagement and user adoption carry equal weight with technical delivery. Revision exercise (suggestion): For a hypothetical operational workflow, map the users, pain point, success criteria, data dependencies and UAT questions; practise explaining the proposed solution without jargon.
RAG and structured LLM outputs
Posting: Working knowledge of RAG, prompt engineering and structured outputs is essential. Revision exercise (suggestion): Explain how ingestion, chunking, embeddings, vector indexing and retrieval affect answer quality, then define an output schema and how you would handle an invalid response.
Agent orchestration and human review
Posting: The role applies tool-use, multi-agent and autonomous-workflow patterns, with human-in-the-loop review. Revision exercise (suggestion): Sketch an agent workflow and mark where tool calls, expert approval and escalation should occur; explain why each boundary is needed.
Document data and enterprise integration
Posting: The role assesses data quality and schemas, processes OCR and multimodal documents, and integrates with enterprise systems. Revision exercise (suggestion): Trace a PDF or image from extraction through validation to an API integration, noting missing fields, schema mismatches and dependencies to flag early.
AI evaluation and deployment controls
Posting: Responsibilities include accuracy, hallucination and retrieval evaluation, regression testing, pre-deployment failure analysis and anomaly escalation. Revision exercise (suggestion): Draft a small test matrix covering normal, edge and adversarial inputs, with a metric or review method and an escalation trigger for each.

Skills to revise

Build a small Python RAG pipeline
Posting: Python proficiency and RAG pipeline development are essential. Practice exercise (suggestion): In Python, ingest a small set of sample documents, compare two chunking choices, index embeddings in a vector database and inspect retrieved passages for several questions.
Implement an agent with a review gate
Posting: Agent configuration, tool-use, structured outputs and human-in-the-loop interfaces are part of the role. Practice exercise (suggestion): Prototype a tool-using agent that returns a defined output structure and pauses for human review before a consequential action; test malformed outputs and tool failures.
Process a mixed-format document
Posting: OCR, PDF/image extraction and multimodal pre-processing are specified. Practice exercise (suggestion): Extract text from a sample PDF containing an image, record extraction errors, and produce clean, traceable text segments for downstream retrieval.
Validate an integration boundary
Posting: The role validates schemas and integrates solutions with enterprise systems; basic REST/async API and microservices understanding is required. Practice exercise (suggestion): Define a sample request and response schema, implement validation at a mock API boundary, and document how missing fields and upstream failures are handled.
Run repeatable AI quality tests
Posting: Evaluation, regression testing and structured agent testing precede deployment. Practice exercise (suggestion): Assemble a small question set with expected evidence, score retrieval relevance and unsupported answers, then rerun it after a prompt or retrieval change.
Facilitate UAT and document handover
Posting: The engineer facilitates UAT, captures feedback and hands solutions to operational owners with guides and walkthroughs. Practice exercise (suggestion): Rehearse a short demo for a non-technical user, capture findings in an issue log, and write a one-page guide covering intended use, review steps and known limitations.

Your preparation checklist

Prepare a concept-to-output project walkthrough
The posting asks for at least one AI or ML project taken to a working output. Prepare a short walkthrough of the problem, your Python implementation, decisions, tests and result; production deployment is advantageous, not required.
Practice translating an operational need
The role involves working with non-technical POSH teams. Draft an example discovery conversation and a one-page specification showing the workflow, user need, proposed prototype, assumptions and scope.
Map data and integration dependencies
Prepare a sample data-readiness checklist covering quality, schemas, access assumptions and enterprise-system dependencies. Be ready to explain how you would flag a blocker to data owners rather than build on an unverified assumption.
Sketch a document-to-answer pipeline
Draw a design covering PDF or image extraction, OCR where needed, chunking, embeddings, vector indexing, retrieval, prompts and structured outputs. Note how you would compare retrieval choices and where an API or cloud service would connect; Azure is preferred, not required.
Design an agent with human review
Prepare a small design exercise showing tool use, orchestration and a human-in-the-loop review point. Explain what the agent may do autonomously, what a domain expert must approve and how feedback changes the workflow.
Create an evaluation and UAT plan
Draft a compact test set and scorecard for answer accuracy, hallucinations, retrieval quality, regression, edge cases and adversarial inputs. Include a UAT feedback log that records gaps and the next iteration.
Prepare a handover example
Outline the configuration, data schema, integration notes, user guide and walkthrough you would provide to operational owners. Include how you would record assumptions and escalate anomalies or platform gaps.
Write questions for the employer
Ask how POSH prioritises use cases, what enterprise-platform and data constraints shape prototypes, and who reviews outputs and owns solutions after handover. These are preparation questions, not claims about its processes.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in Singapore. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…