Role mission
The Data Scientist - Applied AI turns business problems, enterprise data and domain knowledge into working AI solutions. The role combines data science, machine learning and hands-on generative AI development, with responsibility from problem framing and data exploration through prototype, evaluation and pilot support.
This is not a reporting-only analytics role and not a research-only position. The successful candidate must understand the basic principles and limitations of large language models, write production-minded code, build testable prototypes, evaluate performance on real cases, and work with Product, Business, Data and Engineering teams to move useful solutions toward operational deployment.
Why this role exists
The company has growing demand for AI across marketing, consumer intelligence, e-commerce, CRM, supply chain and corporate functions. Converting that demand into reliable products requires more than model access. It requires clean and well-understood data, appropriate model and architecture choices, rigorous evaluation, failure analysis and practical development capability.
This role provides the hands-on scientific and technical depth needed by AI Use Case Squads. It helps the company distinguish promising demonstrations from solutions that can perform reliably in real workflows.
Key responsibilities
1. Frame business and data problems
- Work with AI Product Leads, FDEs, Business Owners and Domain Experts to translate business needs into testable data and AI problems.
- Define the target decision, user, workflow, output, quality standard and measurable baseline before selecting a model or technique.
- Identify the available data, labels, knowledge sources, permissions, biases, gaps and observability constraints.
- Determine whether the problem is best addressed through analytics, traditional machine learning, generative AI, deterministic rules or a combination.
2. Prepare and analyze data
- Acquire, clean, transform and explore structured and unstructured data using Python and SQL.
- Build reproducible datasets and feature or document-processing pipelines for experimentation and evaluation.
- Perform exploratory analysis to identify patterns, data-quality issues, leakage, selection bias and meaningful business segments.
- Define representative train, validation and evaluation samples that reflect real workflow variation and exceptions.
- Document data lineage, assumptions, definitions and known limitations.
3. Develop generative AI and machine-learning solutions
- Build hands-on prototypes and MVP components using approved models, APIs, open-source libraries and enterprise AI platforms.
- Develop retrieval-augmented generation, classification, extraction, summarization, recommendation, forecasting or other relevant solutions.
- Design prompts, structured outputs, retrieval logic, tool use, guardrails and human-review paths where appropriate.
- Compare model, prompt, retrieval and workflow alternatives using evidence rather than public benchmarks alone.
- Write readable, modular and testable code that can be reviewed and extended by engineering teams.
- Expose prototype capabilities through scripts, APIs, services or workflow components when needed for pilot integration.
4. Evaluate quality, risk and business usefulness
- Create evaluation sets from representative real cases with clear expected outcomes or expert scoring criteria.
- Measure task quality, consistency, groundedness, retrieval performance, latency, cost and failure rates.
- Analyze hallucination, missing evidence, model drift, data bias, prompt sensitivity and low-confidence cases.
- Compare AI-assisted performance with the current workflow and quantify the effect of human review and exception handling.
- Define acceptance thresholds, monitoring metrics and stop conditions with Product, Business and Risk partners.
5. Support pilots and operational deployment
- Work as a hands-on member of cross-functional AI Use Case Squads.
- Partner with Data Engineers, Software Engineers and AI Platform teams to move validated prototypes toward secure, maintainable deployment.
- Contribute code, tests, data contracts, API specifications, model configurations and technical documentation.
- Diagnose production or pilot issues using logs, traces, user feedback and evaluation results.
- Help define fallback, manual takeover, versioning and regression-testing mechanisms.
- Make clear trade-offs among quality, speed, cost, complexity and risk.
6. Build reusable capabilities
- Package reusable data pipelines, retrieval components, evaluation datasets, scoring methods and implementation patterns.
- Share practical guidance with AI Product Leads, AI Builders and Domain Experts.
- Track relevant model and tooling developments and test them against company tasks before recommending adoption.
- Contribute field evidence that improves platform, architecture and AI portfolio decisions.
Large language model knowledge expected
The candidate does not need to be a foundation-model researcher, but must be able to explain and apply the following concepts:
- Transformer and attention intuition, tokens, embeddings, context windows and inference behavior.
- The difference among pre-training, instruction tuning, fine-tuning and in-context learning.
- Sampling, temperature, non-determinism, hallucination and common failure modes.
- Retrieval-augmented generation, chunking, embedding search, reranking, grounding and source citation.
- When to use prompting, RAG, tools, traditional ML or fine-tuning, and the trade-offs among them.
- Structured outputs, function or tool calling, agent workflows and human-in-the-loop controls.
- Offline and online evaluation, golden datasets, model-based evaluation and expert review.
- Data privacy, prompt injection, access control, sensitive-data handling and responsible AI considerations.
