AI Scientist

Metropolitan Commercial Bank•New York, NY

3d•$130,000 - $200,000•Hybrid

About The Position

Metropolitan Commercial Bank is seeking a VP-level Applied AI & Machine Learning Scientist to design, build, and validate production-grade AI/ML and Generative AI solutions in a highly regulated banking environment. This role focuses on high-impact use cases—fraud detection, AML alert optimization, AI-assisted credit memo generation for underwriting decision support, contact center AI assistant/copilots, and personalization for treasury/commercial clients—delivered with rigorous governance, explainability, fairness testing, privacy-by-design, cybersecurity, and model lifecycle controls aligned to SR 11-7 and MCB’s Trustworthy & Responsible AI Principles. The role emphasizes Snowflake as the primary ML platform (e.g., Snowpark Python, UDFs/UDTFs, Tasks/Streams, and Snowflake-native ML). Standard 4-day in-office requirement, 1 day remote (of your choosing)

Requirements

6+ years of relevant work experience.
Expertise in Python (pandas, scikit‑learn), deep learning (PyTorch/TensorFlow), NLP/LLMs, LangChain, embeddings/vector search, and classic ML.
MLOps proficiency with CI/CD, containerization (Docker), registries, and observability; cloud ML (Snowflakes-native ML, Azure ML or Databricks preferred).
Snowflake‑native ML proficiency: Snowpark Python, UDFs/UDTFs, Tasks/Streams; ability to build and operate ML workflows inside Snowflake.
Data engineering competency (SQL, ETL/pipelines, Spark/PySpark); ability to work with structured/unstructured data.
Explainability (e.g., SHAP) and fairness testing; ability to produce interpretable reason codes for ECOA/Reg B adverse actions as applicable.
Strong grasp of SR 11‑7 lifecycle, model documentation, and operational monitoring within three lines of defense governance.
Excellent communication; ability to translate technical detail to business/risk stakeholders and drive decisions.
Curiosity and problem‑solving mindset; ability to balance innovation with disciplined risk management.
Ability to work in a constantly evolving environment
Must have excellent written and verbal communication skills
Must be a good listener and good teacher
Demonstrate analytical, troubleshooting and problem-solving skills
The ability to learn new technologies quickly
Self-directed individual with technology and communication skills.
Ability to take in multiple sources of information with an understanding of the bigger picture need, want, and operation of the Bank.
Collaborative team-player who can find creative and practical solutions in a dynamic work environment.
Ability to handle ambiguity, juggle multiple matters at once, and quickly and seamlessly shift from one situation or task to another.

Nice To Haves

Master’s or PhD in a relevant field (Computer Science, Machine Learning, Data Science, Statistics, etc.) is strongly preferred, especially with research or thesis work related to AI/ML, NLP, or model interpretability.
Financial services domain experience (fraud risk, AML, underwriting, or commercial/treasury analytics).
Hands-on with Snowflake ML/Snowpark (Python), Tasks/Streams, secure external functions; experience with feature management/registry tooling a plus. model registry and pipeline orchestration; Kubernetes a plus.
RAG architectures, vector databases, prompt engineering, and LLM evaluation (accuracy, hallucination, safety).
Fairness toolkits and XAI frameworks; experience preparing models for validation, audit, or regulatory exam discussions.
Familiarity with SR 23‑4 (third‑party risk), NYC Local Law 144 (AEDT), NYDFS Part 500 (cyber).

Responsibilities

Design and implement models for fraud detection, AML alert scoring/triage, AI-generated credit memo drafting and underwriting decision support, contact center AI assistants, and personalization for commercial/treasury use cases.
Leverage modern methods: Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), embeddings and vector databases, transformers, boosting, anomaly/outlier detection, and classical ML.
Embed explainability (e.g., SHAP, interpretable scorecards/monotonic models) and conduct pre-/post-deployment bias testing with documented remediation.
Produce audit-ready documentation (methodology, assumptions, data lineage, limitations, testing) and register models in the inventory with owners/materiality.
Facilitate independent validation/effective challenge; obtain required approvals before deployment; maintain change management and periodic review cadence.
Define monitoring, drift thresholds, retraining triggers, and safe rollback/kill-switch procedures; maintain human-in-the-loop checkpoints for high-impact decisions.
Package, deploy, and operate models via CI/CD, containerization, and model registry; instrument KPIs/KRIs and alerting dashboards. Operate models natively on Snowflake using Snowpark Python, UDFs/UDTFs, Tasks/Streams, and secure external access where required.
Partner with Engineering to integrate models via secure APIs/batch; ensure scalability, resiliency, and observability in cloud/on‑prem (e.g., Snowflake, Azure ML, Databricks).
Design for ECOA/Reg B (adverse action specificity), UDAAP, FCRA, GLBA privacy, and NYDFS 23 NYCRR 500 cybersecurity requirements.
Apply privacy-by-design (data minimization, purpose limitation, retention), strong access controls/segregation, and secure SDLC/red teaming for GenAI stacks.
Support due diligence, testing, and ongoing monitoring of vendor AI/data providers per SR 23‑4; evaluate conceptual soundness, fairness, and security.
Negotiate/verify contractual controls (no vendor training on MCB/NPI, subprocessors disclosure, audit rights, exit/portability).
Ensure AEDT compliance (NYC Local Law 144) for any HR-related AI tools.
Collaborate with Model Risk, Compliance/Legal, Cyber/IT, Data Privacy, Internal Audit, and business owners to meet objectives while staying within risk appetite.
Communicate complex results, risks, and limitations clearly to technical and non‑technical stakeholders (management committees, examiners).
Evaluate emerging ML/GenAI methods, LLM evaluation techniques, Snowflake‑native capabilities (e.g., vector search, orchestration), and governance tooling; lead POCs within established control gates.
Mentor junior staff; promote responsible AI practices, documentation standards, and reproducibility.