• Own the AI execution and infrastructure layer used by research and product teams • Design and build high-performance Python systems for: - scalable model inference - training orchestration - large-scale data processing • Partner closely with research and backend engineers to productionize models and expose them via APIs • Design and operate distributed pipelines and task queues for batch and streaming workloads • Optimize GPU inference for latency, throughput, and cost efficiency • Own the MLOps lifecycle, model deployment and versioning; monitoring and alerting • Build and maintain CI/CD pipelines for services and ML workflows • Debug and resolve performance bottlenecks across Python, GPUs, networking, and storage • Contribute to infrastructure design decisions and long-term architecture

• Design system architectures involving LLMs, intelligent agents, and multi-agent systems; • Develop end-to-end Generative AI pipelines, from experimentation to production deployment; • Build autonomous agents, custom tools, functions, and integrations with external APIs; • Implement orchestration between agents, memory systems, expanded context windows, and decision-making mechanisms; • Conduct evaluation, monitoring, and observability of models (RAG, agents, prompts, and behaviors); • Optimize cost, latency, and quality of production solutions; • Work with prompt versioning, automated testing, and continuous validation; • Collaborate with product, marketing, journalism, analytics, and data engineering teams to evolve the company’s Generative AI platform.

• O foco é a implementação de fluxos de raciocínio autônomo, integração via MCP e controle de estado não-determinístico. • **Orquestração de Fluxos Agentivos:** Construir lógicas de loops (Pensar, Agir, Observar) para tarefas complexas de Supply Chain (análise de rupturas, recalcular Sell-In) utilizando frameworks como **LangGraph, CrewAI, Agno ou Bedrock AgentCore**. • **Implementação de Servidores MCP:** Desenvolver e expor ferramentas via *Model Context Protocol*, criando pontes seguras (APIs) entre LLMs e bases **PostgreSQL, Iceberg** e Motores de Regras. • **Engenharia de Raciocínio e Estado:** Gerenciar memória de curto/longo prazo e implementar **Thought Signatures** para garantir coesão em execuções multi-passos (Gemini 3 e Claude 3.5). • **Otimização de Tool/Function Calling:** Desenhar contratos estritos de **JSON Schemas**, implementando lógicas de fallback e retentativas para mitigar alucinações em chamadas externas. • **Disciplina de AgentOps:** Garantir a observabilidade total da IA (tracing, latência, consumo de tokens e caminhos de decisão) via **LangSmith, Phoenix ou AWS CloudWatch**. • **Guardrails e Segurança:** Configurar barreiras automatizadas (ex: **Amazon Bedrock Guardrails**) para interceptar alucinações e bloquear ações não autorizadas em dados mestres.

• Conduct experiments with LLMs and explore different architectures and techniques such as SFT, RLHF, Adapters, LoRA, and related approaches to improve model capabilities • Develop and maintain evaluation methodologies and frameworks to measure model quality, accuracy, and user satisfaction using both offline and online metrics • Optimize models for inference efficiency, speed, and scalability to meet production requirements • Collaborate closely with data scientists, software engineers, and product managers to integrate AI solutions into the product pipeline