AI Engineer

Overview

Build the core engineering primitives of ServiceTitan's Agent OS, the shared runtime, context, memory, action, trust, and evaluation platform behind role-specific AI agents for contractors.

What you'll do

• Build the agent runtime for tool use, delegation, pause/resume, durable checkpoints, retries and failure recovery, so agents resume safely without losing state or duplicating side effects.
• Build typed tools and action contracts: governed reads, proposed writes, precondition checks, business invariants, scoped permissions, idempotency, audit trails and rollback.
• Build tenant-scoped context and memory systems: context assembly, retrieval, freshness controls, provenance, transcripts, artifacts, tool results and replayable evidence.
• Build trust and approval infrastructure: human-in-the-loop gates, approval thresholds, reversibility, tenant policy enforcement and audit history for financial, contractual, dispatch, warranty and compliance-sensitive workflows.
• Build evaluation and observability: offline and online evals, scenario libraries, simulation, trajectory review, regression detection, cost and latency telemetry and autonomy promotion gates.
• Help product teams package prompts, tools, context requirements, policies, evals, rollout controls, ownership and rollback into governed reusable capabilities.
• Contribute to model and inference architecture tradeoffs across latency, cost, quality, structured outputs, caching, fallback behavior, provider choice and model routing.
• Write production code and reliable APIs, workflows and services for agent execution, and review teammates' implementations.
• Inspect traces, debug failures and improve production behavior, including stale context, wrong tool calls, missing permissions, unsafe actions, poor retrieval, latency spikes and cost regressions.
• Design evaluation scenarios and regression suites for agent workflows.
• Partner with domain teams to turn agent use cases into reusable platform patterns and define platform contracts for tools, actions, approvals, context, memory, evidence and evaluation.

What you'll need

• B.S., M.S. or PhD (or equivalent practical experience) in Computer Science, Engineering, Physics or a related field.
• 0-3 years of professional, internship or substantial project experience building software.
• Strong hands-on coding ability in Python (strongly preferred) or a comparable backend language such as TypeScript, C# or Java, with clean, tested code.
• Practical understanding of LLM application basics: prompt and context assembly, retrieval, tool calling, structured outputs and agent workflows.
• Working knowledge of SQL and databases, REST APIs, Git and async or background processing.
• Production-safety instincts: thinking about failure modes, validation, permissions, idempotency, audit trails and monitoring, even if applied mainly in projects.
• Good engineering judgment and clear written and verbal communication.

Nice to have

• Experience building AI, ML, data, platform, infrastructure, workflow, automation or developer-platform systems in production.
• Experience with modern LLM architecture: model gateways, memory, human approval patterns.
• Exposure to distributed or event-driven systems: queues, durable execution, message-driven architectures.
• Experience designing or operating evaluation systems: behavioral evals, regression suites, scenario tests, trajectory review, simulation, online metrics or production monitoring.
• Data and context instincts: unstructured data, vector search, metadata, provenance, freshness, source authority and privacy boundaries.
• Experience with databases, warehouses or search systems such as PostgreSQL, SQL Server, Snowflake, BigQuery, Elasticsearch or vector stores.
• Experience building services on public cloud infrastructure such as Azure, AWS or GCP.
• Awareness of reliability, security, observability and multi-tenant SaaS constraints.

Details

• Work mode: remote in the US.
• Small, senior AI platform team, partnering closely with Product, Architecture, Security, Data Platform, Atlas, and domain engineering teams.