New opportunity

Applied Scientist II, Risk and Compliance Solutions (RCS)

Amazon · Bengaluru, India

About this role

We are looking for an Applied Scientist to join the Data and Analytics team within Risk and Compliance Solutions (RCS). Our team builds the data, analytics, and modeling layer that RCS runs on, from regulatory reports and filings to the intelligence that tells Compliance Officers and our first line of defense partners where risk concentrates and which controls hold. You will build machine learning and generative artificial intelligence systems in that layer, take them into production, and own the evaluation science that lets Compliance rely on them.

Responsibilities

Formulate ambiguous compliance problems as ML and GenAI problems with Compliance and 1LoD partners, and choose methods that fit the data, the risk, and the audit standard
Develop risk models that learn from rare, delayed, and noisy labels to score which risks and control weaknesses convert to breaches, and calibrate them so a score carries a defensible meaning
Build GenAI systems that remove manual effort from requirement translation, control testing, and regulatory reporting, with human approval gates at every stage
Design the evaluation for every ML and GenAI system you ship: golden sets that compliance experts label, metrics suited to rare outcomes and generated text, and release criteria a system meets before Compliance relies on it
Deploy and monitor models in production with software development engineers, including drift detection, performance tracking against realized outcomes, and retraining criteria
Explain model decisions, quantify uncertainty, and document methodology so Compliance Officers, auditors, and examiners can trace how a system reached each output
Create reusable science components, such as feature sets, scoring services, and evaluation harnesses, so an approach that works in financial crime compliance extends to the next domain without a rebuild
Review peer models for statistical soundness, mentor data scientists and engineers on ML methods and experimental design, and publish or patent where the work and confidentiality allow

Required qualifications

3+ years of building models for business application experience
Experience programming in Java, C++, Python or related language
Experience in any of the following areas: algorithms and data structures, parsing, numerical optimization, data mining, parallel and distributed computing, high-performance computing

Preferred qualifications

Experience in patents or publications at top-tier peer-reviewed conferences or journals
Experience in financial modeling, or experience designing and implementing controls or experience performing audits over financial reporting environment
Experience with training and deploying machine learning systems to solve large-scale optimizations, or experience with data infrastructures: relational analytic DBMS, Elastic-Search, and Big Data EMR/EC2/Glue/Lambda

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

Skills for this role

Machine LearningGenerative AIPythonJavaC++Amazon SageMakerAmazon BedrockAmazon RedshiftAWS LambdaAlgorithms and Data StructuresData MiningDistributed Computing

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

Building Rigorous Evaluation Frameworks for ML and GenAI
A core priority is designing evaluation science for every model and agent shipped, including labeled golden sets, domain-specific metrics for generated text and rare outcomes, and strict release criteria that ensure compliance teams can safely rely on automated outputs.
Developing Risk Models for Rare and Delayed Labels
The role requires building predictive risk models that learn from sparse, noisy, and delayed signals to score which open risks and control weaknesses convert into breaches, with careful calibration for defensible decision-making.
Automating Compliance Workflows with Generative AI
Another key focus is creating GenAI systems with human approval gates to eliminate manual effort across regulatory requirement translation, control testing, and reporting.
Productionizing and Scaling Reusable Components
You must take models from formulation to production using AWS infrastructure like SageMaker and Bedrock, while creating reusable science components and feature sets so that solutions in financial crime compliance extend smoothly across other compliance domains.

Likely challenges to think through

Managing Rare and Noisy Labels in Risk Modeling
Training models when confirmed compliance breaches are rare and arrive months later creates significant data sparsity and noise. To prepare, review techniques for semi-supervised learning, anomaly detection, extreme class imbalance, and score calibration under delayed feedback loops.
Satisfying Rigorous Audit and Regulatory Standards
Every model output faces internal audit and external regulatory examination, treating auditor scrutiny as a core design requirement. Preparation should involve mastering model interpretability methods, uncertainty quantification, and thorough methodological documentation so decisions can be fully traced.

Practice your answers

Technical practice

Handling Rare, Delayed, and Noisy Labels in Risk Modeling

What to coverDiscuss strategies for training classification or regression models when confirmed breach labels arrive months after an event and exhibit extreme class imbalance. Address techniques like semi-supervised learning, synthetic data generation, robust loss functions, and probability calibration to ensure scores carry defensible meanings.

Designing Evaluation Frameworks for Generative AI Systems

What to coverExplain how you would design an evaluation harness for a GenAI system used in regulatory requirement translation or control testing. Discuss the creation of expert-labeled golden sets, metrics suitable for evaluating generated text and rare outcomes, and rigorous release criteria before compliance teams rely on outputs.

Model Interpretability and Uncertainty Quantification for Auditors

What to coverDescribe methods for explaining machine learning model decisions and quantifying prediction uncertainty so that internal auditors and external examiners can trace outputs. Discuss feature attribution methods, confidence intervals, and maintaining thorough methodological documentation.

Production Deployment and Drift Detection on AWS

What to coverOutline an architecture using Amazon SageMaker, AWS Lambda, and Amazon Redshift to deploy, monitor, and maintain risk models in production. Discuss how you would implement data drift detection, performance tracking against realized outcomes, and automated retraining criteria with software development engineers.

Creating Reusable Science Components Across Domains

What to coverExplain how you would design modular feature sets, scoring services, and evaluation harnesses so that an approach built for financial crime compliance can extend to other domains without a full rebuild. Discuss software design principles for scalable machine learning components.

Parsing and Processing Unstructured Regulatory Documents

What to coverDiscuss algorithmic approaches and data structures for parsing large volumes of complex regulatory text or filings efficiently. Address text extraction, data mining, and numerical optimization techniques relevant to handling large-scale corporate data.

Parallel and Distributed Computing for Large-Scale Optimization

What to coverExplain how you approach distributed computing or high-performance computing challenges when training large models or processing massive datasets on cloud infrastructure. Discuss parallelization strategies and memory management for large-scale data mining.

Reviewing Peer Models for Statistical Soundness

What to coverDescribe your approach to peer reviewing statistical models for soundness, bias, and experimental design flaws. Discuss how you evaluate assumptions, cross-validation strategies, and potential overfitting in complex business applications.

Behavioural practice

Translating Ambiguous Business Problems into Science Solutions

What to coverDescribe a situation where you had to collaborate with non-technical stakeholders, such as compliance or legal partners, to frame an ambiguous problem into a concrete machine learning formulation. Explain how you aligned data methods with operational risks and audit standards using the situation, action, and result framework.

Balancing Model Performance with Strict Regulatory Constraints

What to coverShare an experience where you had to modify a high-performing model because it failed to meet interpretability, auditability, or fairness standards. Detail how you evaluated the trade-off between predictive accuracy and compliance requirements, and what the final outcome was.

Building Consensus on Evaluation Criteria for New Technology

What to coverProvide an example of a time when you established release criteria or evaluation metrics for a novel AI or machine learning system where no prior benchmark existed. Explain how you worked with domain experts to label golden sets and gain stakeholder trust.

Mentoring Engineers and Reviewing Peer Models

What to coverDescribe an instance where you mentored junior data scientists or software engineers on experimental design or statistical rigor, or where you conducted a thorough peer review of a critical model. Detail how you provided constructive feedback and improved team output.

Sharpen your knowledge

Topics to prepare

Evaluation Science for ML and Generative AI
Conceptualize methodologies for building golden labeling sets, metrics for generated text and rare outcomes, and release criteria. Exercise: Design a comprehensive evaluation harness for a text-to-requirement GenAI model with human approval gates. Relevance: Directly addresses the requirement to own evaluation science so Compliance teams can rely on shipped systems.
Risk Modeling under Rare, Delayed, and Noisy Labels
Study statistical methods for handling extreme class imbalance and delayed feedback loops in predictive risk scoring. Exercise: Simulate a breach-risk model using delayed labels and test score calibration techniques to ensure defensible meanings. Relevance: Directly tied to developing risk models that score control weaknesses converting to breaches.
Model Explainability and Auditability Standards
Examine methods for quantifying uncertainty and explaining model decisions for regulatory scrutiny. Exercise: Document a model methodology and decision path using feature attribution techniques suitable for internal auditors and examiners. Relevance: Aligns with the core expectation that an auditor's question is treated as a design requirement.
Algorithms, Data Structures, and Numerical Optimization
Review foundational computer science principles and optimization algorithms for business applications. Exercise: Implement and optimize a parsing or data mining algorithm in Python or Java to handle large volumes of regulatory text efficiently. Relevance: Matches the explicit qualification in algorithms, data structures, parsing, and numerical optimization.

Skills to revise

Python, Java, or C++ Programming
Practice writing clean, production-ready code in at least one of the required programming languages. Exercise: Write a modular scoring service component with unit tests and logging in Python or Java. Relevance: Fulfills the basic qualification for programming and writing production-quality code alongside models.
AWS Cloud and Machine Learning Infrastructure
Gain hands-on familiarity with cloud tools for deploying and monitoring machine learning workloads. Exercise: Set up a pipeline skeleton using Amazon SageMaker and AWS Lambda for inference and performance tracking. Relevance: Directly supports the stack and the requirement to deploy and monitor models in production with software development engineers.

Your preparation checklist

Review Core Programming and Data Structures Fundamentals
Refresh proficiency in Python, Java, or C++ with a focus on algorithms, parsing, and data mining structures. Completion looks like writing clean, modular code snippets for data ingestion and text preprocessing tasks.
Prepare Risk Modeling and Rare-Label Strategies
Study techniques for handling class imbalance, delayed feedback loops, and score calibration. Completion looks like drafting a conceptual approach for scoring risks when confirmed breach labels are sparse and delayed.
Design a Generative AI Evaluation Framework
Formulate an evaluation strategy for GenAI systems used in compliance testing or document translation. Completion looks like documenting how you would construct expert-labeled golden sets and define release criteria before deployment.
Map Out Production Deployment and Monitoring on AWS
Familiarize yourself with deploying and tracking machine learning workloads using Amazon SageMaker, AWS Lambda, and Amazon Redshift. Completion looks like sketching an end-to-end inference and drift-detection pipeline architecture.
Prepare Model Interpretability and Documentation Artifacts
Examine methodologies for quantifying uncertainty and explaining model decisions for auditors. Completion looks like writing a sample model methodology document that breaks down feature attribution and decision traceability.
Formulate Targeted Questions for the Employer
Draft thoughtful questions regarding how compliance officers interact with human-in-the-loop approval gates, how evaluation sets are maintained across domains, and how the team collaborates with software development engineers.

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 India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…