Senior AI Engineer & Tech Lead — Brasília, Brazil · Remote worldwide
Multi-agent orchestration and RAG in Python (LangGraph · CrewAI) for regulated sectors — FinTech, GovTech, LegalTech, Telecom and Health. Inference and model serving with Rust, vLLM and NVIDIA Triton on Kubernetes and on-premise GPU. Nine years taking mission-critical systems from prototype to production under federal audit scrutiny.
Hiring through the consultancy (B2B): ortzion.com
Three layers, in a deliberate order of priority: the agentic layer is the product, inference is the engine, the platform is the factory floor.
Agentic systems in Python with LangGraph and CrewAI: deterministic routing, typed tool contracts, explicit autonomy boundaries and retrieval grounded in corporate knowledge. Auditable cognitive platforms — without exposing the business to hallucination risk.
Model serving with vLLM and NVIDIA Triton, high-concurrency APIs in Rust (Tokio · Axum) and gRPC, and self-managed GPU infrastructure keeping latency and cost per token under control — including where public cloud is not an option.
Kubernetes (k3s · Rancher), CI/CD, MLflow and observability with Langfuse and OpenTelemetry. Delivery paths that turn prototypes into production workloads with data lineage, continuous evaluation and audit-defensible behaviour.
Through Ortzion — an independent AI engineering consultancy with an on-premise-first practice: self-hosted Kubernetes (k3s) and bare-metal GPU for workloads that cannot move to public cloud. Available as a company (B2B) or directly as an individual. Meet the consultancy: ortzion.com
AI-agent platform for financial workflows: Python polyrepo on FastAPI/Granian, orchestration in LangGraph and CrewAI, retrieval on Milvus and pgvector, inference through a Triton gRPC client. Performance-critical paths moved to Rust.
Clinical decision-support backend combining data-science pipelines with LLM agents and RAG over medical knowledge sources. Typed tool-use contracts for deterministic, auditable outputs — non-negotiable in regulated health software.
Research framework for autonomous decision-making under high uncertainty: multi-agent reinforcement learning (MARL), just-in-time logistics optimisation for tactical scenarios, computer vision and telemetry.
Results by domain — full details, periods and scope live in the CV and on LinkedIn.
Technical lead of the multi-agent ecosystem handling customer interactions and commercial flows: LangGraph/CrewAI over a hexagonal architecture, two-tier contextual memory (Redis + MongoDB), A2A/JSON-RPC integration and Langfuse tracing as the governance baseline. Agent template adopted as the convention across squads.
ML pipeline (XGBoost · BERTimbau · Polars) over a 44M+ record federal eSocial dataset, reconciling ~8.79M misclassified records with full decision lineage — every automated classification traceable and defensible before the federal audit court (TCU).
NLP over unstructured court cases, an agentic core with deterministic routing and secure RAG across case law and internal policy. GPU deployments on Kubernetes via Rancher with zero-trust identity on Keycloak.
Intelligence layer for contract classification isolated from the transactional core; anomaly-detection refactor cutting risk-analysis time by 40% and provisioning automation removing 90% of operational overhead on AWS.
High-performance predictive APIs (FastAPI · gRPC) serving transactional microservices at peak, secure gateways internalising LLM capabilities and financial RAG on pgvector/Milvus — isolated, audited and governed under Central Bank and LGPD requirements.
End-to-end individual delivery of the tax-reform CST-IBS/CBS module — scope originally sized for three engineers over ~13 weeks — with production go-live in May 2026, bitemporal modelling on IBM DB2 and director-level recognition.
Roots in mission-critical banking: deploy cycle cut from ~3 days to ~2 hours on the SISAM platform, migration of MUMPS/Caché monoliths to Spring Boot/Oracle microservices, and nearly a decade of Java, Angular and Kubernetes in regulated environments.
Universidade Cruzeiro do Sul · in progress
Universidade Cruzeiro do Sul · in progress
Universidade Cruzeiro do Sul · 2021 – 2023
Universidade Cruzeiro do Sul · 2020 – 2023
UDF Centro Universitário · 2015 – 2018