Staff AI Engineer

Description

The company builds legal data intelligence software, including a cloud platform that helps law firms, corporations and government agencies organize large volumes of legal information and surface what matters for litigation, investigations and regulatory work.

The role sits in the Discovery group, which builds Document Review, aiR for Case Strategy, Intelligent Fact Gen, Memo Gen, Proof Charting, Translate and Redact. These systems work over very large legal document sets, and their output has to hold up before counsel, opposing counsel and a court.

This is a hands-on staff role spanning the whole Discovery portfolio, with ownership of how these systems are built, evaluated and run. It is not a research role and not a management role.

The hybrid/remote posting type is listed.

Responsibilities

• Set technical direction across several teams and build the hardest pieces alongside them, from prototype through production. Decide which designs to hand off and which to build yourself. You will not own one team's backlog.
• Own the full inference path: retrieval, context building, model selection and routing, agent orchestration, tool interfaces, structured output, graceful fallback, and observability that lets you debug non-deterministic failures quickly.
• Define how quality is measured. Build golden sets, offline test suites, regression gates in CI and online quality monitoring. Every change to a prompt, model or retrieval approach is measured before it ships.
• Own token cost and latency targets. Track cost and latency for each workflow, set unit targets, and use data to decide routing, caching and model tier.
• Design capability logic so it is separate from how it is delivered, behind clear interface contracts. One capability should be deployable as a service, an MCP skill or an embedded feature without a rewrite.
• Design the data and retrieval layer: batch and streaming pipelines, indexing and embedding approach, hybrid search, tenant isolation, and lineage that traces each answer to its source document.
• Build for defensibility: accurate citations, privilege and confidentiality controls, human review paths, and the audit trail customers need to defend a workflow if challenged.
• Detect quality drift, rising hallucination or refusal rates, and creeping latency or cost before customers do. Lead root cause analysis, prevent repeats, and share findings across Discovery.
• Raise the standard across teams through design reviews, reference implementations, documented paths from proof of concept to production, and mentoring of senior and lead engineers.
• Work with Applied Science, Product, Design and other domains, and drive decisions to completion without formal authority. Bring coding agents and modern development practice into the teams while keeping the engineering bar high.

Requirements

• 7+ years building and running large-scale distributed cloud systems.
• 3+ years shipping production systems built on models that real customers rely on. Prototypes and demos do not count.
• Deep knowledge of LLMs, RAG, agentic workflows, prompt and context engineering, vector and hybrid search, orchestration frameworks, and observability for non-deterministic systems.
• You have run an evaluation practice: you built the eval sets, metrics and gates that showed a team whether a change helped or hurt.
• Accountable for architecture across cloud services and data pipelines in multi-tenant SaaS, covering scalability, reliability and security.
• Strong quantitative skills for model behavior: eval metrics, error analysis, and trade-offs between accuracy, latency, throughput and cost.
• Technical leadership across teams without formal authority, and the writing to present a decision, its alternatives and trade-offs to executives and get a clear answer.
• Preferred (optional): experience in legal technology, eDiscovery, search, document intelligence or another field where output must be defensible and auditable.
• Preferred (optional): agentic and multi-agent systems, including merging specialized agents into one generalist design without losing task quality.
• Preferred (optional): MCP servers or model-facing tools and skill interfaces.
• Preferred (optional): reducing inference cost at scale through routing, distillation, caching or smaller fine-tuned models.
• Preferred (optional): governance, monitoring and traceability practices in regulated or high-trust settings.
• Preferred (optional): Azure, Kubernetes, CosmosDB, and progressive delivery through feature flags.

Conditions

Hybrid or remote work.

Salary range is $174,000 to $262,000. The final offer depends on experience, skills, qualifications and internal pay equity. Offers near the top of the range are uncommon, leaving room for later salary growth.

Total compensation includes a competitive base salary, an annual performance bonus and long-term incentives.

The company is committed to fair and equitable pay practices.

Required skills include agentic AI, algorithms, automation, debugging, distributed systems, evaluation metrics, LLM architecture, LLM evaluation, LLMOps, LLM orchestration, machine learning operations, model evaluation, performance tuning, problem solving, project management, RAG pipelines, software development, system design and technical leadership.