Principal AI Engineer (LATAM Remote)

Overview UP.Labs is a dynamic venture studio dedicated to building innovative startup companies from the ground up. Our team thrives on solving complex problems, driving technological advancements, and creating impactful digital products. We're seeking a highly skilled professional to join our growing team and contribute to our mission of launching the next wave of successful startups. Technical Challenge We're building an agentic system as the core delivery mechanism for one of our ventures, and we're looking for a production agent-systems engineer who will own and grow the agentic function end-to-end — not a notebook builder or an ML researcher. You'll design the architecture for agents across the business, ground them in a complex proprietary domain through strong context and retrieval engineering, and build systems that take real action for real users, not just chatbots. Responsibilities Own and grow the agentic function end-to-end: set architecture across internal (data-engineering & data-science agents) and external (customer/partner-facing agents, MCP servers) surfaces, make build-vs-buy calls, and grow a team under you as we scale. Design, build, and deploy production LLM agents that take consequential action (write-back, execute changes) with human-in-the-loop controls — including tool interfaces (MCP, function calling), tiered tool access, approval gates, and ro [... source excerpt omitted ...] d vector stores where relevant, without owning ML model training. Communicate agent architecture, trade-offs, and roadmap to execs and investors; act as a player-coach who writes production code today while owning the function's direction and hiring as the team grows. Required Skills 5+ years shipping production software; strong full-stack/backend engineering (Python core, TS/Node or Go a plus), including production APIs, data models, testing, and CI. Proven track record building and shipping LLM agents to production with real users — multi-step, tool-calling, stateful, with orchestration (LangGraph or equivalent) — and the ability to explain the control loop, not just the fram [... source excerpt omitted ...] rience building eval harnesses for agentic systems (e.g. MLflow, LangSmith, or custom) with fluency in determinism, drift, and guardrails. Experience owning a retrieval system in production, including chunking vs. structured retrieval trade-offs and evaluating retrieval quality. Experience shipping agents that take consequential, real-world action, with a clear point of view on approval/guardrail/rollback architecture (bonus: an incident where the agent did the wrong thing, and how it was handled). Track record leading an agentic initiative or team end-to-end — from architecture to production — including communicating agent systems to both execs and investors. Preferred Skills