AI Agent Engineer

Company Description

Coru is transforming finance in Latin America with AI, building trust, empathy and inclusion through Adi, the intelligent assistant that revolutionizes the sales and customer service experience for enterprise financial institutions. We give our partners' users humanized, hyper-personalized digital experiences while driving greater Revenue Generation and Operational Efficiency for our partners.

Coru's ADI Platform is the enterprise layer that operates conversational agents over PSTN voice, WhatsApp Business and web widgets, with full governance over prompts, tools and behavior — built for the regulatory and operational realities of each market we serve. Today, more than 400 projects have been delivered with a 95% client satisfaction rate, and over 70 companies trust Coru across Latin America.

Learn more about us at https://coru.com/en/ and https://www.linkedin.com/products/coru-br-adi-assistente-digital-inteligente/

Join us to evoke empathy in every interaction with the consumer.

The role

We're not looking for someone to hand-write features for each client. We're looking for someone who takes a process a person performs today, turns it into an explicit procedure, and transforms it into a reliable agent — with human approval gates and evidence that it works before granting it more autonomy.

This is an architecture role, not an execution one: you design the agent, you don't operate the process.

We welcome candidates across different seniority levels (from semi-senior to lead) — the scope of the role adjusts to your experience.

What you'll do

• Map the SOP (standard operating procedure) of an existing process together with the person who runs it today — a mandatory step before automating it.
• Design agents that execute that SOP over a shared source of truth (not isolated bots that don't integrate with one another).
• Define each agent's human approval gates: what the agent decides, what needs a person's sign-off, and with what evidence.
• Instrument evals and quality metrics (e.g., approval-without-correction rate) that determine when an agent earns more autonomy.
• Iterate on agents already in production: detect degradation, correct behavior, and adjust the SOP when the real process changes.
• Work hand in hand with each process's human owner (finance, compliance, delivery, commercial) — you design and maintain the agent, they're accountable for the outcome.
• Document each agent so it's auditable: what it does, what data it uses, what decisions it made, and why.
• Build the underlying stack: the microservices and APIs behind these agents in Node.js and TypeScript, integrating LLMs (OpenAI, Gemini, Ollama) via frameworks like LangChain, vector databases (Weaviate, Redis) for semantic search and context, and channels such as WhatsApp, Messenger and Web — with automated deployments and continuous monitoring on AWS, alongside QA and DevOps.

What we're looking for

• Experience building systems with LLMs in production (not just prototypes): prompting, function calling, RAG, agent orchestration.
• Ability to translate an ambiguous process into explicit, measurable steps — this is more an exercise in process analysis than in pure programming.
• Judgment to design evals: knowing what to measure in order to trust (or not) an agent's autonomy.
• Comfort working with shared state and event-driven architectures, rather than point-to-point integrations.
• Solid backend engineering with Node.js and TypeScript, microservices architecture, and engineering best practices (version control, documentation, automated testing, code review).
• Sensitivity to the regulatory context: understanding why, in a product that sells to banks, traceability and human gates are not a detail but a requirement.
• Autonomy to work remotely and in direct contact with process owners who are not technical.
• Non-negotiable soft skills: proactivity, critical thinking, and a hands-on, resolving profile.
• Portuguese and Spanish, to work closely with Coru's teams and clients across the region.

How we work

• SOP first: if a process can't be written down, it's not ready to be turned into an agent.
• Full gating at the start: every new agent begins with human approval on each action.
• Autonomy is earned with evidence, not enthusiasm.
• Every agent has a human owner with a first and last name — you design the agent, but you don't replace that accountability.

Nice to have (not required)

• Prior experience in fintech, compliance or financial processes.
• English for technical documentation.
• Prior experience with conversational AI products (chatbots, digital assistants) or no-code/low-code platforms.
• Previous experience in startups or environments requiring management of multiple fronts and fast learning.

Do you have what it takes to be a Coru?

We firmly believe that people are what matters most at Coru, which is why it's essential that you have a strong cultural fit with the whole team. These are the pillars of our culture:

• We are all leaders
• Never give up
• The only problem we can't solve is the one we can't talk about
• All together: diversity of stories and experiences. Different perspectives bring broader horizons