Senior Generative AI Engineer

Cleary Gottlieb Steen & Hamilton LLP•New York, NY

1d•$200,000 - $240,000•Remote

About The Position

Cleary Gottlieb is seeking a Senior Generative AI Engineer to join their AI Acceleration Team. This role involves building bespoke AI solutions for legal work, operating at the intersection of enterprise-grade software and advanced AI. The team is remote-first, collaborative, and values growth, honesty, and curiosity. The engineer will own and advance AI systems, working with engineers, data scientists, and legal domain experts. While a legal background is not required, a genuine interest in legal work is essential. This is a hands-on role focused on building practical solutions for lawyers, not academic research.

Requirements

At least 3 years of professional experience building and deploying AI systems, with at least 1-2 years focused on LLM/GenAI applications in production (not just prototypes or research).
Deep experience with document-heavy NLP: extraction from complex layouts (PDFs, tables, scanned documents), entity recognition, and structured output generation.
Proven ability to design and optimise RAG pipelines and prompt architectures for accuracy, cost, and latency in production.
Strong Python engineering skills.
Familiarity with at least one LLM orchestration framework (LangGraph, LlamaIndex, or equivalent) and at least one vector database (Pinecone, Weaviate, pgvector, or equivalent).
Experience with cloud platforms (AWS or Azure preferred) for deploying and monitoring LLM-backed services, including CI/CD, containerisation, and observability tooling.
Experience designing systematic evaluation methodologies for generative AI (automated evals, golden test sets, faithfulness/hallucination metrics).
Ability to own a workstream end-to-end: scope it, build it, evaluate it, ship it, and clearly communicate tradeoffs and results to non-technical stakeholders.

Nice To Haves

Experience in a startup or high-velocity AI team where you shipped frequently and wore multiple hats.
Master's or PhD in Computer Science, Computational Linguistics, Mathematics, or a related quantitative field.
Vector databases, retrieval systems, or knowledge graphs experience.
Familiarity with model-serving infrastructure (vLLM, TGI, Triton) and GPU-aware deployment.
Domain experience in legal tech, compliance tech, or other regulated industries where accuracy and auditability are non-negotiable.
Published research or significant open-source contributions in NLP, information retrieval, or generative AI.
Experience with knowledge graphs, ontologies, or semantic reasoning over structured legal data.
Experience with Spark or Databricks and related technologies.
Experience with document automation or familiarity with legal workflows.

Responsibilities

Own the full lifecycle of LLM-powered products, from rapid prototyping through production deployment.
Design, ship, and operate AI systems that extract intelligence from complex legal documents.
Orchestrate multi-step agent workflows.
Deliver measurable accuracy improvements under real-world latency and cost constraints.
Design, build, deploy, and monitor LLM-powered document analysis pipelines (extraction, classification, risk flagging) that serve lawyers daily, meeting defined latency SLAs and accuracy benchmarks.
Transform legal data into structured, high-quality datasets that power AI systems.
Architect multi-step, multi-agent systems for complex legal tasks (e.g. due diligence, contract review, regulatory analysis) with robust state management, tool calling, and human-in-the-loop checkpoints.
Build golden test sets, automated eval pipelines, and regression suites.
Implement guardrails (prompt firewalls, output filters, PII redaction) to ensure safety and regulatory readiness.
Manage token budgets across model providers, implement caching and batching strategies, and make data-driven build-vs-buy decisions on model selection (API-served, open-source vs. proprietary).
Translate ambiguous legal workflows into structured AI problems, define acceptance criteria with domain experts, and iterate based on user feedback and eval results.