About Matheus
I am an AI & Data Engineer based in Brasília, Brazil. I build agents and the execution, evaluation, data, and cloud systems behind them.
My path started in hardware. I studied Electrical and Electronics Engineering at Universidade de Brasília and began my career in 2018 building sensor-acquisition circuits and ESP32 firmware. As an electronic engineer, I then designed an embedded energy-monitoring system with LoRa connectivity, hardware and firmware, and a Raspberry Pi gateway over MQTT.
A specialization in Artificial Intelligence moved me into data science in 2019. I built natural-language processing and recommendation applications for banking, then MLOps pipelines and real-time inference services for recommendation and time-series applications in e-commerce, deployed on Kubernetes.
Putting models into production led me to data engineering. As a contractor, I built Google Cloud data pipelines for a home-improvement retailer in Brazil and reduced its recurring BigQuery cost by 25%. For a healthcare company in the United States, I built Airflow workflows and the API layer for a Kubernetes-based machine learning platform.
From 2021 to 2025, I worked on large-scale data platforms in retail e-commerce, growing from data engineer specialist to technology lead. I built change-data-capture automation, a shared Python library over PySpark and Airflow, and Kubernetes operators for Spark, then led the data platform team, owning its architecture, modeling BigQuery with dbt, and moving Spark workloads onto Kubernetes.
Since 2026 I have focused on AI agent solutions as an AI Data Engineer in healthtech. I build and run the AI platform: an LLM gateway, self-hosted LLM observability, agent evaluation, and production agents on Kubernetes, including a production agent harness around the Pi SDK. I also designed Smart Agent Factory, a custom Python graph and durable runtime that turns a reviewed request into a validated agent, with human approval at each gate.
AI engineering
As an AI & Data Engineer, I build agents, execution tooling, and the data and cloud infrastructure that supports them.
I evolved the agent harness's Pi-based runtime in TypeScript, connecting declarative runbooks and deterministic context preparation to governed tools, persistent sessions, execution policy, recovery, tracing, user feedback, and model evaluation.
Read case studyI connect OpenTelemetry and Langfuse traces, BigQuery feedback, model comparison, judge calibration, and instrumentation checks. The resulting loop uses production evidence to guide controlled changes to models, prompts, tools, and policy.
Read case studyI designed and implemented a custom Python graph and runtime with append-only state, checkpoints, leases, deduplication, transactional outbox delivery, budgets, bounded retries, human gates, isolated worktrees, external validation, time travel, and receipts.
Read case studyI build the infrastructure around AI services, including Terraform modules, Kubernetes deployment configuration, access controls, observability, and backup automation. I also create incremental pipelines, analytical models, and data-quality workflows with Python, BigQuery, Dataform, and Dataplex.
Read case studyAgent systems knowledge
I treat an agent as a governed software system. The harness controls the context, tools, state, budgets, evidence, and recovery around the model.
See this approach in a delivered systemI map how a request becomes a specification, plan, implementation, and verified result before choosing an agent topology. The manual decisions reveal where automation is safe and where human judgment is still required.
Planning, implementation, and validation run with clear authority boundaries. They exchange durable artifacts, so each stage receives the context it needs without inheriting an unbounded conversation.
Approved goals become sealed contracts with allowed change scope, runnable evaluations, hidden holdouts, and evidence receipts. A deterministic gate checks the work before model-based review begins.
Risk lanes, human decisions, budgets, retries, and resumable state determine how far a workflow can proceed on its own. Autonomy expands only when the system can show why its result should be trusted.
Reference architecture
The model reasons inside a harness that controls context, tools, state, permissions, budgets, human decisions, and acceptance.
Harness / control plane
Engineering capabilities
The disciplines I connect when building and operating AI systems.
Tool-using agents, multi-agent workflows, and retrieval-augmented generation. Execution controls, model evaluation, tracing, and feedback are part of the engineering work.
LangGraph, LangChain, Pydantic AI, Pi agent harness, LiteLLM, Langfuse, Claude Code
Ingestion, batch and streaming pipelines, analytical models, lakehouses, and data quality. The data foundations that support analytics and AI applications.
Spark, Airflow, Kafka, BigQuery, dbt, Dataform, Delta Lake, Apache Iceberg, Dataplex
Backend services, APIs, execution tooling, and reusable libraries. Typed interfaces, state management, validation, and failure handling connect agents to other systems.
Python, Go, TypeScript, SQL, FastAPI, Pydantic, SQLAlchemy
Infrastructure as code, containers, Kubernetes, CI/CD, and access controls, with monitoring and observability for AI and data workloads.
GCP, Kubernetes, Terraform, GitHub Actions, OpenTelemetry, Prometheus, Grafana
Experience
Projects, technical contributions, and engineering disciplines across my career.
AI Data Engineer
Build AI agents and execution tooling, evaluation and feedback pipelines, and reusable developer infrastructure across Python, TypeScript, data systems, and cloud services.
AI & Platform Consultant
Designed and built an end-to-end multi-agent delivery platform with specialized roles, human decision gates, artifact-based verification, resumable sessions, and isolated Git worktrees.
View case studyData Engineer Specialist to Technology Lead
Led technical direction across multiple data platform teams, supported shared lakehouse and orchestration capabilities, and worked on a Spark-to-Kubernetes migration that improved infrastructure efficiency.
View case studyKubernetes Developer
Built Airflow workflows for data acquisition, model training, and validation, plus the API layer for a Kubernetes-based machine learning platform.
View case studySenior Data Engineer
Optimized BigQuery queries, table design, capacity, and storage placement, reducing recurring monthly BigQuery cost by 25%.
View case studyData Science Specialist
Built MLOps pipelines and online inference services for recommendation and time-series applications, using Python, Kubernetes, Kubeflow, and automated delivery.
Data Scientist
Developed natural-language processing and recommendation applications, exposed models through Python inference APIs, and deployed microservices on Kubernetes.
Electronic Engineer and IT Trainee
Designed an embedded energy-monitoring system with LoRa connectivity, hardware and firmware, and a Raspberry Pi gateway using MQTT. Built sensor acquisition systems with ESP32 and supporting applications.
Credentials
IESB
Universidade de Brasília
The Linux Foundation
The Linux Foundation
HashiCorp
Contact
Tell me about your project and the challenge you are working through.