AI Engineer, Agentic Core

Hace 18 horas

rosario, santa fe, Argentina Medra Jornada completa
Agentic Core
· Full-time# AI Engineer, Agentic CoreRemote
· Argentina / LATAMSemi-Senior / Senior3+ years
* Python
* LangGraph
* LangSmith
* TypeScript
* Postgres
* AWSMany heads, one body. That is our architecture, and it is your job.
Agentic Core owns everything LLM-related at Medra: conversational agents, deep agents for clients and internal teams, and the shared intelligence layer underneath them. It is the difference between an agent that solves a customer's problem and one that gets in the way.## What you will do
* Build multi-agent orchestration
- supervisor patterns, expert sub-agents, routing and state.
* Make agents reliable at hundreds of thousands of conversations across 19 countries.
* Push on evaluation: knowing whether an agent got better is central to how we ship.
* Design the seams between the framework and client work, so shared patterns graduate into Core.
* Shape model and provider strategy. We use several and we swap them.## What makes this hard
* Agents act, not just answer. The system has to know when to act, when to ask, and when to stop.
* A non-deterministic stack. Traces, evals, guardrails and fallbacks
- small changes create surprising behaviour.
* Reusable, not one-off. Client learnings have to become shared capabilities, not a pile of hacks.
* Scale exposes weak architecture. Brittle state or slow tools get expensive fast at this volume.## What we are looking for
* 3+ years shipping production software, with serious LLM work in the recent part of it.
* Strong Python or TypeScript.
* You have built agentic systems
- tool-calling, multi-step workflows, autonomous decisions. Not a RAG demo.
* You have operated a non-deterministic system in production and have opinions on debugging one.
* Spanish fluency. You will write prompts and evaluations in Spanish.
* Comfortable in ambiguity. No specs, just problems.## Nice to have
* Evaluation frameworks
- LangSmith or anything else.
* Voice agents, telephony, latency work.
* WhatsApp Business API or messaging platforms at scale.
* Multi-tenant thinking: something other engineers configure rather than fork.## The process1. *01*30 minutes with our CTO2. *02*Technical interview3. *03*Offer