New opportunity

Software Engineer, Generative AI & ML

Apple · Singapore

About this role

Join Apple's Customer Systems – Gen AI Solutions team in Singapore to build conversational, agentic LLM applications and frameworks that support customers across multiple lines of business. The role contributes hands-on to a multimodal, multi-agent platform, incorporating new models and advances to improve latency, cost and customer experience.

Responsibilities

Contribute to model development and fine-tuning workflows for generative AI features. Design and evaluate retrieval strategies that ground large models in product-relevant data.
Prototype and benchmark multi-round agentic solutions for conversational, knowledge-intensive tasks. Work with data and platform engineers on scalable deployment and monitoring, and participate in architecture discussions.
Collaborate with product and business stakeholders on customer-experience requirements. Support localization and rollout across multiple markets, while keeping current with LLMs, vector search, RAG and LLM agents.

Minimum qualifications:

5+ years of hands-on experience in ML, backend engineering or data engineering, plus 1–2 years of hands-on experience training, fine-tuning or evaluating LLMs.
Foundational knowledge of RAG and vector-based retrieval; exposure to multi-agent orchestration frameworks in Rust and Python; familiarity with PyTorch, TensorFlow or JAX; and proficiency with machine learning libraries such as transformers and datasets.
Experience with data preprocessing, tokenization, pipeline automation and deployment to cloud environments such as AWS or GCP. Strong problem-solving, collaboration and adaptability to production-grade systems are required.
Experience training, fine-tuning or evaluating Chinese-language LLMs for the Greater China market, and fluency in written and spoken Chinese. Experience with multimodal LLMs for voice capabilities, or with speech-to-text or text-to-speech systems, is a plus.

Preferred qualifications

Strong experience partnering with business and engineering teams to deliver AI solutions; a bachelor's or master's degree in Computer Science, Machine Learning or a related field, or equivalent practical experience.

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

Skills for this role

Generative AIMachine LearningLarge Language ModelsLLM fine-tuningLLM evaluationRetrieval-Augmented GenerationVector searchMulti-agent orchestrationConversational AIMultimodal AIPyTorchTensorFlowJAXRustPythonData preprocessingTokenizationPipeline automationTransformersDatasetsAWSGCPSpeech-to-textText-to-speechChineseCloud deployment

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

Grounded LLM features
The role contributes to model development and fine-tuning workflows, and designs and evaluates retrieval strategies that ground large models in product-relevant data.
Multi-turn agentic applications
The team builds conversational, agentic applications; this engineer prototypes and benchmarks multi-round solutions for knowledge-intensive customer tasks.
Platform performance and operations
The role contributes to a multimodal, multi-agent platform with a stated focus on latency, cost and customer experience, partnering with data and platform engineers on scalable deployment and monitoring.
Cross-functional customer experience
The engineer participates in architecture discussions and works with product and business stakeholders to incorporate customer-experience requirements.
Chinese-language and multi-market delivery
The posting requires experience training, fine-tuning or evaluating Chinese-language LLMs for Greater China and fluency in written and spoken Chinese; the role also supports localization and rollout across markets.

Likely challenges to think through

Balancing response quality against latency and cost
Likely preparation challenge, not a stated problem: newer models or more agent steps may improve answers while slowing responses or increasing cost. Prepare to explain how you would benchmark those trade-offs against customer-experience goals.
Evaluating grounded, multi-round behavior
Likely preparation challenge: a retrieval strategy can surface relevant data yet still lead to weak answers over several turns. Prepare an evaluation approach that distinguishes retrieval quality, grounded response quality and agent behavior.
Moving prototypes into monitored production
Likely preparation challenge: an agentic prototype may behave differently at deployment scale. Prepare to discuss how you would work with data and platform engineers on pipeline reliability, monitoring and architecture decisions.
Maintaining quality across languages and markets
Likely preparation challenge: Chinese-language performance and localized customer experiences may not follow from results in another language or market. Prepare to explain how you would evaluate localized behavior and incorporate stakeholder requirements during rollout.

Practice your answers

Technical practice

How would you approach fine-tuning an LLM for a customer-facing generative AI feature?

What to coverPreparation suggestion: Explain how you would choose training data, establish a baseline and evaluate whether fine-tuning improves the intended task. Discuss trade-offs against using an existing model without fine-tuning, including quality, latency and cost.

How would you make an LLM data-preparation pipeline reliable?

What to coverPreparation suggestion: Walk through preprocessing, tokenization and automation for training or evaluation data. Explain how you would detect data-quality problems and make pipeline outputs reproducible.

How would you design vector retrieval to ground answers in product-relevant data?

What to coverPreparation suggestion: Describe how you would prepare and index content, retrieve relevant passages and supply them to the model. Compare plausible retrieval choices using answer relevance, latency and update needs.

How would you evaluate whether retrieval actually improves grounded answers?

What to coverPreparation suggestion: Compare answers with and without retrieved context on representative knowledge-intensive questions. Discuss how you would assess retrieval relevance, answer support and failures where relevant evidence is missed or misused.

How would you prototype and benchmark a multi-round agentic conversation?

What to coverPreparation suggestion: Outline how the system would preserve useful context across turns and coordinate agents for a knowledge-intensive task. Explain what you would measure, including task success, error recovery, latency and cost.

What would you consider when working with multi-agent orchestration in Rust and Python?

What to coverPreparation suggestion: Describe how you would define agent responsibilities, state and interfaces between components. Discuss trade-offs in implementation and testing without assuming a particular orchestration framework.

How would you prepare an LLM application for scalable cloud deployment and monitoring?

What to coverPreparation suggestion: Explain the handoffs you would make with data and platform engineers and the signals you would monitor after deployment. Address how you would investigate changes in quality, latency or cost; the posting names cloud environments but does not prescribe a specific architecture.

How would you assess a newer model for the existing multimodal, multi-agent platform?

What to coverPreparation suggestion: Propose a benchmark against the current approach that reflects conversational quality, customer experience, latency and cost. Explain how you would decide whether an improvement on one measure justifies a regression on another.

How would you evaluate a Chinese-language LLM feature for the Greater China market?

What to coverPreparation suggestion: Discuss Chinese-language training or fine-tuning data, evaluation examples and review of answer quality for the intended market. Explain how you would distinguish language-quality issues from retrieval or model-behaviour issues.

How would you assess adding voice capabilities to a conversational feature?

What to coverPreparation suggestion: Because multimodal LLM, speech-to-text and text-to-speech experience is listed as a plus, treat this as an optional practice topic. Compare possible approaches and discuss how recognition or synthesis errors could affect multi-turn conversations, latency and evaluation.

Behavioural practice

When have you translated stakeholder needs into an AI or software solution?

What to coverPreparation suggestion: Give a real example using the situation, your actions with product or business stakeholders, and the result. Explain how you resolved ambiguity about the intended user experience without assuming a specific Apple process.

When have you partnered with other engineers to make a system production-ready?

What to coverPreparation suggestion: Describe the situation, your contribution alongside data or platform engineers, and the result. Focus on a concrete deployment, scalability or monitoring decision and how responsibilities were coordinated.

When have you worked through a disagreement about system architecture?

What to coverPreparation suggestion: Use a real situation, the evidence and trade-offs you brought to the discussion, and the outcome. Show how you collaborated with partner teams even if your preferred design was not chosen.

When have you adapted a feature for users in another language or market?

What to coverPreparation suggestion: If you have a relevant example, describe the situation, your actions to understand local needs, and the result; otherwise, prepare a clearly identified analogous example rather than inventing one. Discuss what you learned about validating quality beyond a direct translation.

When have you had to evaluate a new technique under changing requirements?

What to coverPreparation suggestion: Give a real example of the situation, how you learned and tested an unfamiliar approach, and the result. Emphasize how you used evidence to decide whether it was suitable for a production-grade system.

Sharpen your knowledge

Topics to prepare

LLM development and evaluation
The role includes model development, fine-tuning and evaluation. Suggested revision: compare when to fine-tune a model versus evaluate an existing one, and outline how you would test quality before a customer-facing release.
RAG and vector retrieval
The posting calls for grounding large models in product-relevant data. Suggested revision: diagram a RAG architecture and explain how retrieval quality can affect answer quality.
Multi-round agent architecture
The team builds conversational, multi-agent applications. Suggested revision: trace a knowledge-intensive conversation across several turns, identifying where agents exchange context and where failures might occur.
Production trade-offs
The posting emphasizes scalable deployment, monitoring, latency, cost and customer experience. Suggested revision: work through an architecture scenario in which improving answer quality increases latency or cost, and explain how you would assess the trade-off.
Chinese-language and market localization
Chinese-language LLM work for Greater China and written and spoken Chinese are minimum qualifications; the role also supports rollout across markets. Suggested revision: identify evaluation issues that could arise when adapting a conversational feature for Chinese-language customers.
Multimodal voice capabilities
Experience with multimodal LLM voice capabilities or speech-to-text or text-to-speech systems is listed as a plus, not a requirement. If relevant, revise by sketching how voice input and output would fit into a conversational LLM application.

Skills to revise

Prepare model data pipelines
The posting specifies preprocessing, tokenization, pipeline automation and proficiency with transformers and datasets. Suggested exercise: prepare a small text dataset, document cleaning and tokenization decisions, and make the preparation steps repeatable.
Run a Chinese-language LLM evaluation
Hands-on training, fine-tuning or evaluation of Chinese-language LLMs is required. Suggested exercise: define a small Chinese-language conversational test set, evaluate outputs against stated criteria, and explain the errors you find.
Benchmark retrieval strategies
Designing and evaluating retrieval for product-relevant grounding is a stated responsibility. Suggested exercise: compare two vector-retrieval configurations on the same questions and inspect whether the retrieved passages support the resulting answers.
Prototype a multi-round agent flow
The role prototypes and benchmarks multi-round agentic solutions and asks for exposure to orchestration frameworks in Rust and Python. Suggested exercise: build a small conversational flow in a framework you know, then explain its state handling and how you would benchmark it.
Plan deployment and monitoring
Cloud deployment experience is required, and the role partners with data and platform engineers on scaling and monitoring. Suggested exercise: outline a deployment to an environment named in the posting, such as AWS or GCP, and specify what you would monitor for latency, cost and failures.
Translate customer requirements across markets
The role works with product and business stakeholders on customer experience and supports localization. Suggested exercise: write a brief Chinese-language explanation of a proposed conversational feature, then turn a sample customer-experience requirement into a testable evaluation criterion.

Your preparation checklist

Map experience to the stated qualifications
Prepare a one-page evidence sheet covering the posting’s 5+ years in ML, backend or data engineering; 1–2 years training, fine-tuning or evaluating LLMs; and work with preprocessing, tokenization, pipeline automation and cloud deployment. Note specific examples you can substantiate.
Build an LLM workflow walkthrough
Prepare one project walkthrough showing the data, model framework or libraries used, fine-tuning or evaluation approach, results and limitations. The role names PyTorch, TensorFlow or JAX and libraries such as transformers and datasets; discuss only tools you have used.
Explain a retrieval design and its evaluation
Sketch how you would ground a conversational model in product-relevant data using RAG and vector retrieval. Include an example query flow, evaluation measures and likely failure cases; label this as a proposed design, not an Apple system.
Prepare an agentic-system benchmark
Outline a multi-round, knowledge-intensive task and a way to compare agent approaches on response quality, latency and cost. Be ready to explain orchestration choices; the posting specifically mentions exposure to frameworks in Rust and Python.
Rehearse a production architecture discussion
Draw a high-level deployment and monitoring plan for a conversational LLM application. Explain how you would work with data and platform engineers, and ground any cloud examples in your actual AWS, GCP or hybrid deployment experience.
Prepare Greater China and localization evidence
Select an example you can substantiate of training, fine-tuning or evaluating a Chinese-language LLM, and prepare to discuss it in spoken and written Chinese. Add how you would test localized customer experience across markets; treat voice, STT or TTS as optional because the posting calls them a plus.
Draft stakeholder questions
Bring questions about how the team defines customer-experience requirements, evaluates retrieval and multi-round agents, balances latency against cost, and coordinates localization and rollout. These are questions to ask, not claims about Apple’s current 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…