- This position is open to candidates located in Colombia or Costa Rica only -
You'll build new product features on top of LLMs, including agents, retrieval systems, and the workflows around them, and you'll use AI tooling heavily to do it. This is a greenfield role on a small, fast-moving team. Requirements arrive loosely defined, you help define them, and you own what you ship from prototype to production. Our first major milestone lands 60 days out, so you should expect to be contributing code in week one and shipping something real well before your first quarter is up.
What you'll do
- Design and ship user-facing features backed by LLMs, including evaluation, failure handling, and cost management
- Make architectural decisions and defend them, weighing tradeoffs across speed, cost, complexity, and risk
- Build systems that stay fast and available as usage grows, with sensible latency budgets, graceful degradation, and no single points of failure
- Handle PII and financial data responsibly, thinking through what data flows where, what reaches a model provider, and what the failure modes look like
- Build the automation and data pipelines those features depend on
- Own deployment and operation of your services
- Help set the team's direction on AI tooling and approach
Required Qualifications
- Strong Full Stack development experience with JavaScript and TypeScript. Exposure to or experience with Go is a plus. Senior-level software engineering experience. You can design a system, review someone else's code, and debug something you didn't write
- Architectural judgment. You think in systems rather than tickets, and you can explain why you chose one approach over another to both engineers and non-engineers
- Experience assessing risk in environments handling PII or financial data, and a working sense of where the real exposure sits versus where it only looks scary
- Track record building highly available, scalable, performant systems, and the instincts to know when that matters and when it's premature
- Ships production code with AI coding tools such as Cursor or Claude Code as a normal part of the workflow, not just experimentation
- Has built and shipped LLM-backed features to real users, such as agents or RAG pipelines, including the unglamorous parts
- Practical judgment about where AI helps and where it doesn't, grounded in things you've actually built
- Comfortable owning deploys end-to-end on a PaaS like Fly.io or Render, and able to find your way around AWS or GCP when a project needs managed services
- Builds automation and data workflows to support the above
- Moves fast under real deadlines and works well with ambiguous requirements
- Motivated by building new features rather than supporting existing systems
Preferred Qualifications
- AI-assisted code review workflows
- Evals, fine-tuning, or model training experience