Senior AI Engineer - Internal Solutions

What is Cobre, and what do we do?

Cobre is Latin America’s leading instant b2b payments platform. We solve the region’s most complex money movement challenges by building advanced financial infrastructure that enables companies to move money faster, safer, and more efficiently.

We enable instant business payments—local or international, direct or via API—all from a single platform.

Built for fintechs, PSPs, banks, and finance teams that demand speed, control, and efficiency. From real-time payments to automated treasury, we turn complex financial processes into simple experiences.

Cobre is the first platform in Colombia to enable companies to pay both banked and unbanked beneficiaries within the same payment cycle and through a single interface.

We are building the enterprise payments infrastructure of Latin America!

The team you'd join

AI Engineering is the platform domain behind every AI-native product at Cobre: a shared AI toolkit, company-wide MCP servers, and the orchestration, tooling and monitoring standards that other squads build their agents on top of. We don't just consume models — we build and operate the infrastructure that makes it safe, fast and cheap for the rest of the company to do so.

You'd join the AI Engineering team and build AI products for the Internal Solutions, turning Cobre's customer support into an agentic system. Today, a meaningful share of client questions — about integrations, API behavior, transaction status, webhook payloads — still get resolved by a person, in whichever channel the client happens to use, whenever someone is available. Your job is to build an agentic layer, reused across channels, that resolves what it can against Cobre's real systems and escalates cleanly to a specialist in Customer Success or Integrations when they can't.

What we are looking for:

An engineer who can take a customer support agent from "it works in a demo" to a system that holds up under real client traffic, across channels and time zones, without ever letting one client's data leak into another's answer.

What would you be doing:

• Own the agentic support layer end-to-end — from migrating the email automation off ad hoc n8n flows onto a single decision engine, through the multichannel entry points (portal chat, WhatsApp, Slack) that will share the same routing and escalation logic.
• Build the deterministic scaffolding around the nondeterministic core — event-driven pipelines, idempotency, retries and dead-letter queues, typed domain models, and an audit trail that holds up to scrutiny.
• Treat client-data isolation as a design requirement, not a refinement — no response can expose one client's data to another, in any channel, under any phrasing of the question. That constraint shapes the architecture from day one, not after a beta.
• Integrate against Cobre's real sources of truth — documentation, the public API, sandbox, Snowflake and Salesforce — behind clean, well-tested interfaces rather than ad hoc calls.
• Design the escalation path — confidence thresholds and handoff logic so a CS or Integrations specialist picks up exactly when the agent shouldn't keep going, with enough context that the client never has to repeat themselves.
• Instrument before you build — correlation IDs that survive every async hop, dashboards for resolution rate and time-to-first-response, alerts that mean something.
• Keep model and vendor choices behind an anti-corruption layer — so a change of model, provider, or agent framework is a one-adapter change, not a rewrite.
• Work AI-natively and hold that work to the normal bar — AI assistants, MCP servers and agents are part of the daily toolkit here, but anything produced with them gets reviewed, tested and owned exactly like code you wrote by hand.
• Partner across the org — with Customer Success and Integrations on what the agent should actually resolve, and with the rest of AI Engineering so you're reusing shared orchestration and monitoring standards instead of rebuilding them.
• Feed what you learn back into the platform — the patterns, guardrails and reusable components that make the next squad's agent faster to build.

What do you need:

• 5+ years building and operating production backend services, with real ownership of what you shipped after it shipped.
• Depth in at least one backend-focused language, and willingness to work in Go if you don't already — Cobre's stack is Go-first, with Python in the mix.
• Distributed systems fundamentals — asynchronous and event-driven processing, at-least-once delivery, idempotency, eventual consistency, and a real answer for what happens when a downstream provider is down.
• Cloud-native AWS development — Lambda, DynamoDB, S3, SQS/SNS — plus containers, IaC and CI/CD as a normal part of your work, not someone else's job.
• Architectural judgment — hexagonal architecture and DDD in practice: rich domain models, ports and adapters, and the discipline to keep third-party SDKs and vendors at the edge of the system.
• Testing discipline — high coverage as a gate rather than an aspiration, table-driven and contract tests, and an honest answer for how you test a component that isn't deterministic.
• Experience building systems that sit in front of sensitive, multi-tenant data, with the isolation and access-control discipline that requires.
• Working knowledge of LLM-backed features in production — prompt design, tool-calling, retrieval, and a clear-eyed view of where they fail and what the fallback is.
• Comfort in an AI-assisted engineering workflow — using coding assistants and MCP-based tooling, and holding that output to the same review bar as anything else.
• Professional fluency in Spanish, our working language day to day, and working English for technical documentation.
• Strong written and spoken communication, in both.

Nice to have

• Experience with agent-orchestration frameworks (LangGraph or similar) and evaluating multi-step, tool-calling agents.
• Experience building or maintaining MCP servers or comparable tool-exposure layers for LLMs.
• Familiarity with defining golden datasets, eval suites, and confidence thresholds for AI-assisted workflows.
• Kafka, schema registries, and contract evolution in production.
• Experience migrating a product off workflow-automation tools (n8n, Zapier, or similar) onto a centralized service.
• Payments, treasury or fintech background, especially anything touching client integrations or support tooling.