Role Description We're looking for a Senior AI/ML Engineer to design and build the foundational AI layers, multi-step agentic workflows, and developer tooling for the broader engineering organization. You'll focus on the internal runtime for AI: - Implementing prompt frameworks - Enforcing deterministic structured outputs - Deploying production-grade Human-in-the-Loop state automations - Turning raw context into reliable, zero-hallucination agent-based actions Qualifications - Overall Experience: 7+ years in Backend Software Engineering and AI/ML Application Engineering - Agentic Architecture: 2+ years engineering multi-step LLM workflows, autonomous agent loops, and multi-agent orchestration frameworks in production - Developer Platform Enablement: experience building internal SDKs, tools, or structured pipeline abstractions used by other product engineering teams to ship AI features rapidly - Proficiency in Multi-Step Agentic Workflows & Tooling - Proficiency in Advanced Amazon Bedrock & Prompt Engineering - Proficiency in Structured Output Pipelines & Performance Ingestion - Proficiency in State Management, Automation Consistency & HITL - Production fluency in Python - Proficiency in TypeScript/Node.js for Vercel AI SDK and AWS edge workers Requirements - Vercel AI SDK, LangGraph, or LlamaIndex for agentic orchestration - Reliable tool-calling (function-calling) interfaces for LLM-driven internal system navigation - Short-term and long-term agent memory architectures with ultra-low-latency in-memory stores (Amazon MemoryDB / Redis OSS / Valkey) - Amazon Bedrock APIs (foundational model invocation and chaining for agent loops) - Programmatic prompt optimization, few-shot prompting, dynamic context injection, and prompt caching - Amazon Bedrock Guardrails (safety gating, prompt injection mitigation, PII redaction) - Type-safe schema enforcement with Pydantic (Python) or Zod (TypeScript) - Ingestion workers consuming Amazon MSK (Kafka), SQS, and log sources - Optimistic Concurrency Control (OCC) and async write-delegation to prevent PostgreSQL deadlocks - Deterministic multi-step pipelines with AWS Step Functions or LangGraph - Durable HITL workflows (pause prior to DB writes or API transactions, persist state, resume on approval) - Idempotent execution layers for safe agent loop retries without record duplication - Working knowledge of C# (.NET Core) or Java for enterprise pipeline integrations Benefits - Remote work - 13 floating holidays - 15 vacation days per year completed - Good working environment Company Description Every qualified candidate who meets the requirements outlined in the job description will be considered in this hiring process without distinction. Furthermore, Jalasoft is an equal opportunity employer. We wholeheartedly embrace our responsibility to make employment decisions without regard to race, age, marital or social status, national origin, disability, sex, gender identity or expression, or any other characteristic or group of candidates or employees unrelated to their qualifications and suitability for the position. Our management is committed to upholding this policy with respect.

• Mapear oportunidades em crédito e cobrança onde agentes de IA podem substituir ou ampliar processos manuais, considerando fluxos de trabalho, riscos operacionais e critérios de sucesso em ambientes regulados • Projetar e manter arquiteturas de **Retrieval Augmented Generation (RAG)**, alimentando agentes com informações precisas e atualizadas • Gerenciar pipelines de ingestão de dados não estruturados (PDFs, áudios, textos, imagens), incluindo OCR e extração de entidades • Desenvolver e orquestrar sistemas multi-agente utilizando frameworks como **LangGraph, CrewAI ou AutoGen** • Implementar padrões arquiteturais como **ReAct, Plan-and-Execute e Reflexion** • Integrar agentes com APIs internas, sistemas legados e fontes de dados de crédito e cobrança • Estruturar e versionar prompts de sistema, instrução e contexto com técnicas avançadas (**chain-of-thought, few-shot prompting, structured outputs, self-consistency**), garantindo comportamentos determinísticos e auditáveis • Avaliar a qualidade das respostas dos agentes com métricas especializadas (**Faithfulness, Answer Relevancy, Context Precision, Hallucination Rate**) • Implementar **guardrails** e camadas de segurança para prevenção de alucinações, prompt injection e vazamento de dados sensíveis, em conformidade com a LGPD • Operar pipelines de **LLMOps** para monitoramento contínuo de custos, latência, qualidade e drift comportamental dos agentes em produção • Garantir versionamento de prompts, modelos e bases de conhecimento com rastreabilidade completa • Acompanhar o desempenho dos agentes em produção, iterando sobre base de conhecimento, lógica de orquestração e estratégias de fallback • Medir o impacto em KPIs de negócio (redução de tempo de análise de crédito, aumento na taxa de recuperação, melhoria na experiência do cliente) • Elaborar documentação técnica conforme padrões internos e exigências regulatórias, garantindo auditabilidade das decisões automatizadas

• Architect, prototype, and deploy RAG pipelines, combining vector search, hybrid retrieval, reranking and contextual compression techniques. • Contribute to design and orchestration of multi-agent LLM systems using community frameworks and custom orchestration layers. • Work on a variety of information extraction, information storage and information retrieval problems for both structured and unstructured data. • Partner with cross-functional (product, infra, data engineering, and software engineering) to build robust, high-scale systems that underlie all of our data processing and ML Operations.

• Build and operate the LLM and agent infrastructure every team at Coinbase builds on. • Ship high impact AI agents on top of the LLM infrastructure. • Serve as the company's single path to LLMs and the substrate for the full agent lifecycle: build, deploy, run, observe, improve.