Senior ML Engineer

TextUs · Denver, CO · 1 mo ago

HybridEngineering$180/hrFull-time

We're moving from a product where AI is a feature you can turn on to one where it's a layer that runs through everything: response suggestions, abuse detection, summarization, lead scoring, intent classification. That shift only works if there's an engineering layer underneath that treats ML systems with the same rigor as the rest of production.
We're AI-pragmatic, not AI-maximalist. Most of what we ship will run on frontier model APIs with retrieval and good prompt engineering. Some will run on small classifiers we train ourselves. A few things will justify fine-tuning against our eleven years of conversation data. Your job is to know which is which, and to build the platform that lets us move between them without rebuilding from scratch every time.
You own the ML and AI engineering layer end to end.
The ML Ops platform:

Model registry, feature pipelines, and deployment pathways that any engineer in the org can use
Evaluation infrastructure that catches regressions before they hit prod, not after
Drift detection, online evals, cost and latency monitoring
Rollback and progressive rollout patterns built for ML systems, not retrofitted from generic CD

LLM-powered features built on frontier APIs: prompt engineering, retrieval, structured generation
Eval frameworks that tell us whether any of it is actually working
Cost and latency budgets, and the engineering work to stay inside them
Human-in-the-loop feedback loops that make features measurably better over time

Small specialized classifiers where they're the right tool: intent, opt-out, urgency, abuse
Selective fine-tuning when the task, the data, and the economics line up
Inference infrastructure that holds under campaign-volume load

Build-vs-buy calls. Know when a frontier API is the right answer, when a managed service is fine, when to fine-tune, and when a regex would have done the job.
Guardrails so product engineers can ship AI features without becoming ML experts
A clear, defensible point of view on what customer data can be used for what, and how it gets handled

We're an AI-native engineering org. Claude Code is at 100% licensed and roughly 80% active across engineering. You're expected to use it heavily for your own work, and to push the org on where AI changes how ML itself gets built: synthetic eval generation, automated regression detection, faster experimentation loops.
You'll also be the person other engineers come to when they want to add an AI feature to something they own. The bar is that they leave the conversation knowing more than when they walked in.

6+ years of engineering experience with at least 3 years focused on ML platform, ML Ops, or applied ML in production
You've been on call for models. You know what breaks and how to design so it breaks less.
Strong applied LLM experience. You have opinions on eval, RAG, prompt engineering, and where each fails. You can tell the difference between a demo and a production system.
Comfortable in Python across the modern ML stack. Comfortable enough in Ruby on Rails to integrate with our product.
Cloud-native infrastructure depth (AWS preferred). Containers, IaC, the boring parts of running production systems.
Track record of good build-vs-buy decisions. You've said no to building something more often than you've said yes.
Clear communicator. You can explain a model's behavior to a PM and an inference pipeline to a backend engineer in the same afternoon.

Small teams, real ownership. You'll build the ML stack the right way, with vision towards what it should look like for years to come.
AI-native by default. The expectation is that you use Claude Code (and whatever comes next) as part of how you actually work. We invest in the tools and the patterns.
Outcomes over output. We care that the right things ship safely, not that the dashboard looks busy.
We hire for judgment. Tooling will change; the instinct for where a system is going to break should not.

Similar jobs