New opportunity

Senior / Principal Data Scientist, GenAI

AIG PC Global Services, Inc. · Atlanta, United States of America

About this role

AIG seeks a hands-on senior or principal data scientist for its Generative AI team in Atlanta. This individual contributor will architect, develop and deliver enterprise-level GenAI solutions for insurance and risk-management applications, focusing on large language models and machine learning in an agile, production-oriented environment. Team members are expected to work primarily in the office.

Responsibilities

Lead data science and GenAI projects from business requirements through delivery, collaborating with product managers, engineers and business leaders.
Design scalable ML pipelines and retrieval-augmented generation (RAG) frameworks that integrate with enterprise data structures.
Build evaluation frameworks for LLM effectiveness, manage ground-truth dataset quality and use findings to guide product development.
Monitor production solutions for performance, reliability and accuracy; provide technical leadership through code quality and architectural decisions.

Required qualifications

8+ years of engineering experience in a development environment and 6+ years in a data science role emphasizing NLP and ML in an agile, production-oriented setting.
3+ years of practical experience with open-source LLMs such as Llama 3 or Mixtral, including prompt engineering, inference optimization and fine-tuning; and 3+ years building GenAI solutions involving RAG, validation pipelines, observability and monitoring.
Experience delivering multiple GenAI, analytical or ML projects from conception to production; expert Python proficiency, including its data science ecosystem such as PyTorch, Pandas and Scikit-learn.
A master’s degree in data science, computer science or a related quantitative field. Strong communication skills, including explaining technical concepts to nontechnical stakeholders, and the ability to work in a fast-paced environment.

Preferred qualifications include Palantir Foundry or AIP experience or certification; ontology and digital-twin modeling; big-data performance optimization with Apache Spark or PySpark; and agentic architectures for multi-step reasoning or integration with enterprise systems. AIG describes health, wellbeing, financial-security and professional-development benefits, as well as Volunteer Time Off and Matching Grants programs.

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

Skills for this role

Generative AILarge Language ModelsMachine LearningNatural Language ProcessingRetrieval-Augmented GenerationPythonPyTorchPandasScikit-learnLlama 3MixtralPrompt EngineeringInference OptimizationModel Fine-TuningLLM EvaluationGround Truth Dataset ManagementML PipelinesValidation PipelinesObservabilityProduction MonitoringPalantir FoundryPalantir AIPOntologyDigital TwinsApache SparkPySparkAI AgentsCommunicationCross-Functional Collaboration

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

Own delivery from requirements to production
The role leads GenAI and data science solutions from business requirements through delivery as a hands-on technical driver. Prepare a concise account of how you would make architectural decisions, guide implementation and define production readiness.
Architect enterprise RAG and ML pipelines
The posting calls for scalable pipelines and RAG frameworks that integrate with enterprise data structures. Prepare to explain retrieval design, data integration and how you would test the resulting system.
Make LLM evaluation useful for product decisions
The role builds evaluation frameworks, manages ground-truth dataset quality and uses findings to guide the product roadmap. Prepare to distinguish model-quality measures from dataset-quality checks and show how results would inform changes.
Keep deployed solutions reliable and accurate
Production oversight covers performance, reliability and accuracy; the qualifications also emphasize validation, observability and monitoring. Prepare to describe what you would monitor and how you would investigate a decline after release.
Lead technically across functions
This individual contributor partners with product managers, engineers and business leaders while setting direction through code quality and architecture. Prepare to explain technical trade-offs in terms nontechnical stakeholders can use.

Likely challenges to think through

Evaluation rigor versus delivery pace
A likely challenge, not a reported problem: an agile, production-oriented team may need decisions before a ground-truth dataset fully represents real use. Prepare an approach for defining an initial evaluation set, documenting its limits and expanding it as findings emerge.
Retrieval quality versus enterprise-scale integration
A likely design difficulty is preserving useful retrieval as a RAG framework connects to complex enterprise data structures. Prepare to discuss how you would validate retrieved evidence, diagnose failures and decide where pipeline changes matter most.
Model quality versus inference performance
Because the role includes open-source LLMs, inference optimization, fine-tuning and production monitoring, improvements in output quality may compete with serving performance. Prepare to compare prompt, retrieval and model-level changes using both evaluation results and production measurements.

Practice your answers

Technical practice

Suggested practice: How would you design a RAG solution around enterprise data structures?

What to coverThe role calls for scalable RAG frameworks integrated with enterprise data. Explain how you would connect source data to retrieval and generation, then evaluate the trade-offs among retrieval quality, freshness, access controls and latency.

Suggested practice: How would you evaluate whether an LLM solution is effective?

What to coverThe posting assigns this role responsibility for LLM evaluation frameworks that guide product development. Describe task-specific measures, representative test cases and how results would inform a decision to improve, release or reconsider a solution.

Suggested practice: How would you maintain a trustworthy ground-truth dataset?

What to coverGround-truth dataset quality is an explicit responsibility. Discuss how you would define labels, check consistency and coverage, investigate disagreements and prevent evaluation data from becoming stale.

Suggested practice: When would you use prompting rather than fine-tuning an open-source LLM?

What to coverThe posting names open-source LLMs such as Llama 3 and Mixtral, along with prompt engineering and fine-tuning. Compare the approaches in terms of task fit, data needs, evaluation evidence and ongoing maintenance rather than assuming either is always preferable.

Suggested practice: How would you improve LLM inference performance?

What to coverInference optimization is a stated qualification. Explain how you would diagnose latency and resource use, test changes against output quality and choose an approach suitable for a production workload.

Suggested practice: How would you troubleshoot a production GenAI solution whose accuracy has declined?

What to coverThe role monitors production performance, reliability and accuracy and calls for observability and validation pipelines. Walk through signals you would inspect, how you would isolate changes in data, retrieval or model behavior, and how you would verify a fix.

Suggested practice: How would you build a scalable ML pipeline in Python?

What to coverThe posting calls for scalable ML pipelines and expert Python proficiency, citing PyTorch, Pandas and Scikit-learn as examples. Describe data and model validation, reproducible processing, testing and the trade-offs you would assess before selecting libraries.

Suggested practice: How could an ontology support an AI application built on complex enterprise data?

What to coverOntology and mapping real-world entities and relationships into a digital-twin framework are preferred, not required, qualifications. As preparation, explain how you would represent relationships, handle ambiguous mappings and assess whether the model improves an AI application's use of data.

Suggested practice: How would you diagnose a slow large-scale data-processing job?

What to coverBig-data performance optimization with Spark or PySpark is listed as preferred. As preparation, describe how you would identify a bottleneck, test a targeted change and check that the resulting data remains correct.

Suggested practice: When would an agentic architecture be justified?

What to coverMulti-step reasoning and integration with enterprise systems are preferred experience areas. As preparation, compare an agentic approach with a simpler workflow, including how you would validate intermediate steps and limit unreliable actions.

Behavioural practice

Suggested practice: Tell me about taking a GenAI or ML project from requirements to production.

What to coverThe role calls for end-to-end delivery in an agile, production-oriented setting. Use a real situation, your actions in defining scope and resolving delivery trade-offs, and the observed result; distinguish your contribution from the team's.

Suggested practice: Tell me about aligning product, engineering and business stakeholders on a technical approach.

What to coverThe posting emphasizes collaboration across these groups. Give a real situation, how you surfaced differing needs and reached a decision, and what resulted.

Suggested practice: Tell me about explaining an LLM evaluation finding to a nontechnical stakeholder.

What to coverThe role requires communicating complex concepts and using evaluation to guide product development. Describe the situation, how you explained the finding and its limitations, and the decision or outcome that followed.

Suggested practice: Tell me about improving code quality or architecture while remaining hands-on.

What to coverThe posting describes a technical-lead individual contributor responsible for code quality and architectural excellence. Use a real example to show the problem, the changes you personally drove and their observed effect on delivery or maintainability.

Suggested practice: Tell me about responding to a production reliability or accuracy issue under time pressure.

What to coverProduction oversight and the ability to work in a fast-paced, high-stakes environment are stated expectations. Describe the situation, how you prioritized investigation and communicated with others, and the result, including what you learned.

Sharpen your knowledge

Topics to prepare

Open-source LLM trade-offs
Required focus: Llama 3 or Mixtral experience, prompt engineering, inference optimization and fine-tuning. Exercise: compare when you would change a prompt, optimize inference or fine-tune a model, explaining the expected quality and performance trade-offs.
RAG architecture and enterprise data
The role designs scalable retrieval-augmented generation frameworks integrated with enterprise data structures. Exercise: sketch the path from a business question through data retrieval to an LLM response, and explain how the design would handle changing source data.
LLM evaluation and ground-truth quality
The posting calls for evaluation frameworks that measure LLM efficacy and guide product development. Exercise: draft an evaluation plan with representative questions, ground-truth review criteria, failure categories and a decision rule for comparing two versions.
Production ML reliability
The role includes scalable ML pipelines, validation, observability and monitoring for performance, reliability and accuracy. Exercise: walk through a hypothetical production quality drop, distinguishing data, retrieval and model causes and the signals you would inspect.
Ontology and digital-twin modeling
Preferred, not required: mapping real-world entities and relationships into a digital-twin framework. As optional preparation, sketch a small, hypothetical risk-management entity map and explain how its relationships could support an AI application.

Skills to revise

Write and review Python data-science code
Expert Python proficiency is required; PyTorch, Pandas and Scikit-learn are named examples. Exercise: use a small synthetic dataset to clean records with Pandas, build a simple ML baseline with Scikit-learn, and explain the tests and code-quality changes you would request before production.
Experiment with an open-source LLM
Practical work with models such as Llama 3 or Mixtral is required. Exercise: run the same small prompt set through two prompt variants, record response quality and inference performance, and outline a fine-tuning experiment only if the observed failures justify one.
Build a small RAG flow
Building RAG frameworks and validation pipelines is an explicit requirement. Exercise: use synthetic documents to implement retrieval and answer generation, then test cases where the relevant document is missing, outdated or ambiguous.
Evaluate and monitor a GenAI solution
The role builds evaluation frameworks, manages ground-truth quality and monitors production accuracy and reliability. Exercise: label a compact test set, calculate results by failure category, and propose monitoring checks that would reveal a regression after delivery.
Explain an end-to-end technical decision
The posting calls for leading delivery from business requirements to production and explaining complex concepts to nontechnical stakeholders. Exercise: present a concise hypothetical GenAI design to a product manager and an engineer, covering the requirement, architecture, evaluation evidence and remaining risks.

Your preparation checklist

Map the stated qualifications to evidence
Prepare a one-page reference to examples of engineering, production NLP/ML, open-source LLM and GenAI experience; Python work; and the stated master’s degree. The posting specifies experience thresholds, so keep dates and contributions precise.
Build two delivery walkthroughs
Select two GenAI, analytical or ML projects and outline the business requirement, your hands-on architectural and coding decisions, cross-functional collaboration, production release and measured results. The role calls for delivery from conception through production.
Practice an enterprise RAG design
Sketch a hypothetical insurance or risk-management use case, showing how a scalable retrieval and ML pipeline would connect to enterprise data structures. Explain design trade-offs rather than implying this is an existing AIG system.
Prepare an open-source LLM technical example
Bring a concise example using a model such as Llama 3 or Mixtral: explain prompt choices, when fine-tuning was appropriate, inference optimization and relevant Python tooling. These are areas named in the posting.
Draft an evaluation and production-quality plan
Prepare a sample scorecard covering LLM effectiveness, ground-truth dataset quality, validation, observability and monitoring for accuracy, reliability and performance. Explain how findings would inform product decisions.
Separate preferred expertise from core evidence
If applicable, prepare brief examples of Foundry or AIP, ontology and digital-twin modeling, Spark or PySpark optimization, or agentic integration. Label these as preferred qualifications, not requirements.
Rehearse a stakeholder explanation and employer questions
Practice explaining one architectural trade-off to a nontechnical business leader. Prepare questions about the new GenAI team’s priorities, how product and engineering partners define success, and expectations for primarily in-office work in Atlanta.

Availability last checked: Nov 4, 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 United States of America. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…