Applied AI Researcher

Role purpose

Bridge advanced AI research and practical enterprise use cases by validating models, methods, and prototypes that can become production-grade AIRP solutions. The role focuses on measurable business value, rigorous experimentation, model behavior, and safe translation of research into banking-relevant applications.

Client-specific emphasis

• Research must be grounded in enterprise business use cases, not generic AI experimentation.
• Candidates should understand how model, retrieval, data, evaluation, latency, cost, and safety decisions affect production delivery on AIRP.
• Cloud/AWS awareness is valuable because successful research outputs must be handed off to engineering teams building on AWS-hosted AIRP.

Primary ownership

• Applied research agenda for LLMs, NLP, RAG, evaluation, multimodal AI, and agentic workflows relevant to enterprise use cases.
• Prototypes, experiments, benchmark design, model-selection recommendations, and production-readiness evidence.
• Research-to-production handoff with AI engineering, AIRP platform, product, risk, and governance teams.

Key responsibilities

• Conduct applied research in LLMs, GenAI, NLP, information retrieval, multimodal AI, synthetic data, and agentic AI.
• Design experiments to evaluate model performance, robustness, safety, scalability, interpretability, enterprise usefulness, and production feasibility.
• Prototype AI solutions for KYC, credit underwriting, governance tracking, pitch book generation, Banker 360, Customer 360, deal library intelligence, financial crime quality, and sanctions screening.
• Develop evaluation methodologies using golden datasets, adversarial testing, offline benchmarks, human review, business outcome metrics, and risk-specific acceptance criteria.
• Assess prompt optimization, RAG, fine-tuning, instruction tuning, synthetic data generation, distillation, and model adaptation techniques.
• Document model limitations, data assumptions, hallucination patterns, bias risks, performance boundaries, and control recommendations for regulated deployment.
• Collaborate with engineers to convert prototypes into production-ready AIRP requirements, including latency, cost, observability, security, and AWS/cloud deployment considerations.
• Track emerging AI research and translate relevant advances into practical recommendations for the enterprise.

Must-have candidate profile

• Advanced degree preferred, usually MS or PhD in AI, ML, computer science, statistics, computational linguistics, mathematics, or related field.
• Strong foundation in machine learning, deep learning, NLP, transformers, information retrieval, and generative AI.
• Hands-on experience with LLMs, embeddings, RAG, model evaluation, and applied GenAI experimentation.
• Python skills with PyTorch, TensorFlow, Hugging Face, scikit-learn, or equivalent research frameworks.
• Ability to design rigorous experiments and communicate findings to technical, product, business, risk, and governance stakeholders.
• Ability to translate research results into production requirements suitable for an AWS-hosted enterprise platform.

Preferred experience

• Research or applied science experience in banking, finance, compliance, risk, legal, operations, financial crime, sanctions, or enterprise knowledge systems.
• Experience with AWS Bedrock, SageMaker, vector search, MLflow, Databricks, model evaluation tooling, or cloud-based experimentation environments.
• Publications, patents, internal research contributions, open-source AI contributions, or prior research-to-production handoffs.
• Familiarity with Responsible AI, model validation, privacy constraints, audit documentation, and regulated deployment environments.

Initial screening questions

• What research idea did you convert into a prototype or production capability?
• How would you design an evaluation harness for an LLM-based banking use case such as KYC, underwriting, or sanctions screening?
• How do you determine whether fine-tuning, RAG, prompting, or model adaptation is the right approach?
• How do you account for latency, cost, safety, and AWS/cloud deployment constraints in applied research?
• What failure modes did you discover and how did you mitigate them?
• How do you communicate model limitations to non-research stakeholders?