Applied Artificial Intelligence Engineer

You’ll work alongside senior engineers, Product, and our Chief Product & Technology Officer (CPTO) to design, prototype, and ship AI-powered capabilities quicklyThis role is hands-on and iterative, focused on building production-grade agentic workflows that improve how internal communications are curated, designed, delivered, measured, and orchestratedThis is an AI engineering role with real infrastructure ownershipYou won’t be handed a platform - you’ll help build the one our AI features run onIf you like being close to both the model and the metal, this is that jobAgentic Design & Orchestration: Build agentic workflows and orchestration patterns (dynamic routing, tool-using agents, feedback loops) - contributing to the design and owning the implementation of well-scoped componentsEvals & Measurement: Own the eval harness for the features you ship. Build the datasets, the automated grading, and the offline regression suites that tell us whether a prompt or model change actually made things better. Raise the bar for evaluation across the team - judge design, offline and online metrics, and the judgment to know when a number is real. Turn production failures into evals that catch that class of failure next timeInfrastructure for AI: Own work with SRE and build the infrastructure your features depend on, as code. Build the deployment path for AI workloads on EKS alongside our SRE and platform engineersReliability, Cost & Safety: Keep our AI layer trustworthy as it grows. Instrument token spend, latency, and failure rates as first-class metrics. Design for the failure modes that matter - hallucination, timeout, rate limit, cost blowout - and make sure the system degrades gracefully instead of falling over. Respond when things behave unexpectedly in productionPrompts & Model Configuration as Code: Treat prompts as versioned, reviewable, rollback-able artifacts rather than strings someone edited in a console. Own how we move a prompt or model change from idea to production safely, and how we back it out when it turns out worseData Residency & Trust: We run regional deployments because our customers require it. Help make sure AI features respect those boundaries - that customer data stays in-region, model invocations are logged, and what we send to a model is what we intended to send. Think about prompt injection and PII exposure before an auditor doesExecution & Experimentation: Take a defined problem and run with it. Contribute to our quarterly AI roadmap, run experiments to benchmark model value, and iterate systematically on prompts, retrieval, and model configurations based on what the data tells youCross-Functional Partnership: Work closely with Product Management to turn product ideas into concrete technical work - both customer-facing features and internal workflow automationGrow With the Team: Participate actively in design discussions and code reviews. Ask questions early, elevate blockers, and share what you learn. Contribute to documentation practices (agents, markdown, Knowledge Bases) as we figure them out togetherCultural Stewardship: Be a full participant in helping the engineering culture evolve as we growBenefits
Hybrid WorkEmployee Stock OptionsPersonal Development BudgetBi-weekly Lunch & LearnMonthly Team-building ActivitiesHealth & Dental CoverageBonus OpportunitiesExtended Vacation Days (34 days of vacation annually, with summer Fridays, and more)This is a great fit if you’ve shipped LLM features in production, you’re comfortable when the answer involves a Terraform plan rather than a prompt tweak, and you’re ready to go deeper - with senior engineers around you to learn fromYou are able to work collaboratively and independentlyYou have excellent oral and written communication skillsYou’ve written and applied infrastructure as code yourself. Terraform preferred - you’ve authored modules, read a plan carefully before applying it, and recovered from state that didn’t match reality. Terragrunt, CloudFormation, or CDK experience transfersYou default to AI in your own workflow. You’ve already integrated modern AI tools into how you work, and you use them to move faster rather than as a crutchYou bring intellectual humility, empathy, and strong listening skills to the teamYou have familiarity with RAG architectures and vector stores, and an interest in multi-agent orchestration frameworks - you don’t need to have run them at scale, but you should know the landscapeYou can reason about what an AI feature costs to run - tokens, context size, retries, caching - and you treat that as an engineering constraint rather than someone else’s problemYou prioritize a “question first, then answer” approach, thinking critically about problems before committing to a code pathYou hold a Bachelor’s degree (or higher) in Computer Science, Statistics, Mathematics, or Engineering, or equivalent practical experienceYou are comfortable with agile developmentYou’ve taken LLM work past a prototype at least once - you’ve thought about output quality, iterated on prompts against real usage, and dealt with the gap between “works in the notebook” and “works for customers.”You’ve built evaluation beyond trial and error - you can measure accuracy, safety, and latency, and you know when a benchmark win is real and when it’s noiseStrong fullstack experience - Ruby on Rails (or any other backend language/framework) and Vue.Js/React (or any other framework) preferred. You’re comfortable in a production SaaS codebase and can navigate unfamiliar systems without needing everything explained first3+ years of professional software engineering experience, including hands-on work with AI/ML or LLM-powered featuresWorking knowledge of AWS beyond the console: IAM (roles, policies, least privilege), VPC and networking basics, secrets management, and how a containerized workload gets deployed and observed. Kubernetes familiarity is a strong plusYou understand how to structure LLM calls for reliability: using function calling and structured outputs to ground results, testing responses systematically, and designing for failure modes (hallucinations, latency, cost)You’ve built something agentic - an LLM reasoning across multiple steps, making decisions from tool outputs, adapting to feedback. Side projects countYou’ve built an eval harness that changed a decision - where the numbers told you to ship, hold, or roll back, and you listenedYou’ve operated AI in production - you’ve been on the other side of a cost spike, a provider outage, a model deprecation, or a quality regression that only showed up under real trafficYou have a bias toward simplicity - you’re pleased when a well-structured baseline beats something more complex, and you know when not to reach for a model at allYou’ve worked with Amazon Bedrock, or have opinions about model routing and provider abstraction earned the hard wayYou’ve worked within the email ecosystem or a similar high-scale communication platformYou’ve built with AI coding agents and have a view on what they’re good and bad at - we use them, and we’re still figuring out the right practicesYou prototype fast, iterate quickly, and use data to change direction when the evidence says soYou share what you learn with peers - writeups, demos, docs, internal talks
#J-18808-Ljbffr