New opportunity

Algorithm Engineer (Multimodal AI & Trust Intelligence), Marketplace Intelligence & Data (2027 Graduate)

Shopee · Singapore

About this role

Join Shopee’s Marketplace Intelligence & Data team in Singapore to develop AI models for identity verification and AI security. The 2027 Graduate role focuses on computer vision and multimodal AI for production systems.

Responsibilities

Develop and improve models for AIGC detection, media forensics, face intelligence, liveness detection, deepfake defense, document understanding, fraud detection, multimodal reasoning and AI agents.
Participate in data analysis, model training, evaluation and deployment. Design training and inference pipelines using modern deep learning frameworks.
Work with senior engineers and product managers to deliver production-ready solutions, while exploring advances in computer vision, multimodal AI and trustworthy AI.

Requirements

Bachelor’s degree or higher in Computer Science, Artificial Intelligence, Machine Learning or a related field.
Strong Python programming skills and hands-on experience with a deep learning framework such as PyTorch.
A foundation in at least one of computer vision, multimodal AI, vision-language models, large language models, natural language processing or speech processing; familiarity with model training, evaluation and debugging.
Problem-solving ability, curiosity, willingness to learn, communication skills and a collaborative mindset.

Preferred

Image or video analysis experience; familiarity with computer vision or multimodal models; relevant research, competitions or publications; or an AI or algorithm internship.

Shopee offers mentorship from experienced algorithm engineers, work on large-scale production AI systems, exposure to topics including multimodal large models and AIGC detection, and a technical growth path. Applicants with a current right to work in Singapore who do not need Shopee visa sponsorship will be prioritized. Candidates can be considered in only one Sea Group recruitment process at a time.

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

Skills for this role

PythonPyTorchDeep LearningComputer VisionMultimodal AIAIGC DetectionMedia ForensicsFace IntelligenceLiveness DetectionDeepfake DetectionDocument UnderstandingFraud DetectionMultimodal ReasoningAI AgentsModel TrainingModel EvaluationModel DeploymentData AnalysisInference PipelinesModel DebuggingVision-Language ModelsLarge Language ModelsNatural Language ProcessingSpeech ProcessingImage AnalysisVideo AnalysisProblem SolvingCommunicationCollaboration

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

Build models for identity verification and AI security
The role develops computer vision and multimodal models for areas including liveness detection, deepfake defense, AIGC detection, media forensics, and document understanding. Prepare to explain the modeling choices and limitations of one relevant approach in depth.
Work across the model lifecycle
The posting explicitly includes data analysis, training, evaluation, debugging, and deployment. Prepare a concise account of how you would diagnose a weak result and decide what to test next.
Implement training and inference pipelines
Shopee calls for Python, hands-on use of a deep learning framework such as PyTorch, and pipeline implementation. Be ready to explain how you would make a training run reproducible and an inference path testable.
Turn research into production-ready work with others
The role involves exploring computer vision and trustworthy AI techniques while collaborating with senior engineers and product managers to deliver production-ready solutions. Prepare to describe how you would communicate experimental evidence, limitations, and a proposed next step.

Likely challenges to think through

Balancing errors in trust decisions
Preparation hypothesis, not a stated Shopee problem: a liveness or fraud model may need to catch abuse without wrongly rejecting legitimate users. Prepare to discuss evaluation beyond a single accuracy score, including error types and how a decision threshold changes the trade-off.
Handling changing or unfamiliar media
Preparation hypothesis: AIGC and deepfake detection may be difficult when new generation methods or image and video conditions differ from training data. Prepare to explain how you would construct held-out tests, inspect failure cases, and avoid overstating generalization.
Reconciling model quality with deployment constraints
Preparation hypothesis: a stronger multimodal model may also require more inference time or compute. Because the posting calls for both inference pipelines and production-ready solutions, prepare to compare approaches using measured quality and resource costs rather than assuming the largest model is best.

Practice your answers

Technical practice

How would you inspect data before training an identity-verification model?

What to coverThe role includes data analysis and identity-verification problems. Preparation suggestion: discuss label quality, class balance, duplicate examples and train/test separation, including how poor data could distort evaluation.

How would you build a basic model-training workflow in Python and PyTorch?

What to coverThe posting asks for Python, hands-on experience with a deep learning framework such as PyTorch, and training-pipeline design. Preparation suggestion: explain data loading, the training and validation loops, checkpointing and how you would make runs reproducible.

What would you check if a model trains successfully but performs poorly on validation data?

What to coverThe role calls for familiarity with model training, evaluation and debugging. Preparation suggestion: distinguish data or label problems from overfitting and implementation errors, then describe a small, testable check for each.

How would you evaluate a detection model when false positives and false negatives have different costs?

What to coverAIGC detection, fraud detection and identity verification are stated work areas. Preparation suggestion: discuss precision, recall, threshold selection and performance across relevant data slices rather than relying on accuracy alone.

How would you test whether an AIGC detector relies on genuine forensic cues?

What to coverThe job explicitly includes AIGC detection and media forensics. Preparation suggestion: consider held-out generation methods, image or video transformations and spurious dataset cues, and explain what failures would mean for deployment.

How would you approach liveness detection and evaluate its failure cases?

What to coverFace intelligence, liveness detection and deepfake defense are listed model areas. Preparation suggestion: outline a plausible image- or video-based approach, then discuss presentation attacks, changing capture conditions and the trade-off between security and rejecting genuine users.

How would you structure a document-understanding model for a fraud-related task?

What to coverThe posting names document understanding and fraud detection. Preparation suggestion: explain how visual layout and extracted text might contribute, how you would handle unreadable or altered inputs, and what evidence would justify the model's decision.

When would a multimodal approach be more useful than a single-modality model?

What to coverThe role involves computer vision, multimodal models and multimodal reasoning. Preparation suggestion: describe a task where combining visual and text information helps, then address missing inputs, conflicting signals and the added complexity of fusion.

How would you evaluate a multimodal reasoning or AI-agent prototype?

What to coverMultimodal reasoning and AI agent work are named in the job description. Preparation suggestion: define a bounded task and measurable success criteria, then discuss checking unsupported conclusions, error recovery and whether a simpler model would suffice.

What changes when moving a trained vision model into an inference pipeline?

What to coverThe role includes deployment, inference-pipeline design and production-ready AI solutions. Preparation suggestion: cover consistent preprocessing, latency and resource trade-offs, monitoring of input or output changes, and a way to detect regressions.

Behavioural practice

Tell me about a time you worked with people who had different priorities on a technical task.

What to coverThe role involves collaboration with senior engineers and product managers. Preparation suggestion: use a real situation, your actions to clarify priorities or trade-offs, and the result; distinguish your contribution from the team's.

Describe a difficult model or data problem you investigated.

What to coverThe requirements emphasize problem-solving and familiarity with model debugging. Preparation suggestion: explain the situation, the checks you performed and why, and the result, including what remained unresolved if applicable.

Tell me about a time you learned an unfamiliar technique to complete a project.

What to coverThe posting values curiosity, willingness to learn and exploration of new AI techniques. Preparation suggestion: give a real situation, the steps you took to learn and test the technique, and the result rather than just naming resources you used.

Describe a time you communicated an evaluation finding that affected a technical decision.

What to coverThe role includes model evaluation, communication and delivering production-ready solutions with others. Preparation suggestion: state the situation, how you conveyed the evidence and its limitations, and the resulting decision or next step.

Sharpen your knowledge

Topics to prepare

Computer vision or multimodal foundations
The posting requires a good foundation in at least one listed area, such as computer vision, multimodal AI, vision-language models, language models, NLP or speech—not all of them. Choose your strongest area and explain its inputs, model outputs and limitations for an identity-verification example.
Identity verification and AI security
The role names face intelligence, liveness detection, deepfake defense, AIGC detection and media forensics. Compare two of these tasks: define what each detects, what evidence a model could use and where errors could matter. This is a suggested revision exercise, not a stated requirement to know every task.
Document understanding and multimodal reasoning
Both appear among the team's model applications. As a suggested exercise, outline how image and text information might contribute to understanding a document, then identify a case where the two sources conflict.
Model evaluation and debugging concepts
The posting asks for familiarity with basic training, evaluation and debugging. Review how you would separate training from evaluation, recognize overfitting and investigate different kinds of prediction errors; explain your approach using a hypothetical fraud-detection task.
Production model lifecycle
The role spans data analysis through deployment and calls for production-ready solutions. Trace a hypothetical model from data preparation to inference, identifying where you would check data quality and whether predictions still behave as expected after deployment. These checks are preparation suggestions, not specified employer procedures.

Skills to revise

Python model implementation
Solid Python programming is required. Write a small, readable program that loads example labels and predictions, checks their shapes and computes an evaluation summary; explain how you would test an unexpected input.
Deep learning framework practice
Hands-on experience with a framework such as PyTorch is required. In PyTorch or another framework you know, implement a minimal training loop with a forward pass, loss, gradient update and separate evaluation pass; diagnose one deliberate training bug.
Data and error analysis
Data analysis, evaluation and debugging are part of the stated work. Given sample predictions for an image-classification task, group errors by condition, inspect representative failures and propose one targeted change to test. The example task is a practice choice, not a specified project.
Training and inference pipelines
The job involves designing and implementing both pipelines. Sketch and code a small example that applies consistent input processing during training and inference, saves a trained model and runs predictions on new inputs; explain how you would check for preprocessing mismatches.
Technical collaboration
The posting calls for communication and collaboration with senior engineers and product managers. Practice a short explanation of a model result that states the objective, evaluation findings, limitations and a proposed next step without claiming the model is production-ready.

Your preparation checklist

Select a relevant project walkthrough
Prepare a concise account of a computer vision or multimodal project, if available: the problem, data, model choice, your contribution, results and limitations. The role focuses on identity verification and AI security.
Prepare Python and framework evidence
Choose code you can explain that demonstrates Python and hands-on use of a deep learning framework such as PyTorch. Be ready to discuss training decisions and a bug you diagnosed; these match stated requirements.
Practice a model-pipeline design exercise
As interview practice, sketch a training and inference pipeline for one advertised problem, such as liveness detection. Identify inputs, data preparation, model outputs and deployment considerations; the posting does not specify an interview exercise.
Build an evaluation and debugging brief
Prepare a one-page example showing how you assessed a model, examined failure cases and decided what to improve. Connect it to the role’s data analysis, evaluation and deployment responsibilities without inventing project results.
Rehearse a collaboration example
Prepare a short account of how you explained a technical trade-off, incorporated feedback or worked with others. The posting calls for communication and collaboration with senior engineers and product managers.
Draft questions about the team’s work
Ask which identity-verification or AI-security problems the graduate would tackle first, how models are evaluated before deployment, and how engineers and product managers collaborate.
Review application logistics
Note the posting’s Singapore work-rights prioritization and its rule that applicants can be considered in only one Sea Group recruitment process at a time; plan any application questions accordingly.

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…