Brazil · UTC−3English / Portuguese
I’m Matheus Castro, a product-minded software engineer with six years of experience across operational applications, applied computer vision, and production infrastructure.
I turn ambiguous requirements into systems teams can ship, operate, and improve.
Selected work / 01
Confidential work, explained through the decisions and outcomes that can be shared.
For a confidential industrial workflow, I trained and deployed a vision model for counting small components in large batches. The complete camera pipeline ran locally through ONNX at approximately 30 FPS in the measured setup and reached production use across multiple phones.
At CIAg, a private Brazilian science and technology institute, I operate an eight-host on-prem environment spanning core services, Proxmox VMs, GPU compute, networking, GitLab, and deployment tooling. I also turn project architectures into Terraform and CI/CD workflows that development teams can operate.
lower total AWS spend, measured in Cost Explorer, after consolidating smaller EC2 workloads into Dockerized services.
An S3-hosted GitLab backup failed because its archive structure was incompatible after an update. Over a weekend, I repaired the archive, restored the repositories with only nonessential pipeline logs lost, and turned the process into reproducible, real-world-tested recovery scripts.
My commercial work includes extending a legacy logistics workflow, building an application module for a connected agricultural product, and materially contributing to production Kotlin/Spring and NestJS systems. Client identities stay private; the technical scope can still be described clearly.
Capabilities / 02
I’m most useful where a project crosses product decisions, implementation, and the environment that must run it.
New workflows, backend services, and internal tools for established operations. Comfortable with legacy constraints, stakeholder alignment, and incremental delivery.
Kotlin · Spring · TypeScript · NestJS · Java · MySQL
Model experimentation, training, evaluation, and integration—with particular attention to offline inference and whether machine learning is necessary in the first place.
Python · YOLO · ONNX · Anomaly detection
Reproducible cloud and on-prem environments, delivery pipelines, practical quality gates, and handoffs that leave teams able to deploy independently.
AWS · Terraform · GitLab CI/CD · Docker · Proxmox
AWS services used in production
I have also evaluated GCP and OCI for cost and fit; my production depth is in AWS.
Leadership / 03
I proposed and now coordinate a six-person engineering guild at CIAg, with day-to-day responsibility for technical effectiveness, workload continuity, individual support, and leave coordination.
The goal is simple: preserve a direct, trusted channel as the organization grows, so technical and interpersonal constraints can be handled before they become delivery problems.
2019–2020Software Developer · Newcon
2021–nowCIAg · Intern → Research Assistant → Junior Researcher → Technology Researcher
Research / 04
I’m pursuing a Master’s in Computer Science at UNESP. The current work investigates multimodal agents and accessible interaction; I’ll publish the project’s scope, methods, and findings after its formal qualification.
Qualification pending
Community lab / 05
Fast, accessible utilities that process data locally whenever possible.
Create and customize QR codes directly in the browser.
Open tool Local onlyGenerate strong credentials with Web Crypto. Nothing is sent or stored.
Open tool WCAG 2.2Test text and interface colors, then find a nearby passing shade.
Open toolPersonal work / 06
The Stardew Valley projects are small, personal tools made before generative AI entered my workflow. They are here because polish and affection count too.
Notes / 07
Short writing about product decisions, applied AI, infrastructure, accessibility, and the uncomfortable details between them.
Read the notesContact / 08
Tell me what needs to work, who depends on it, and which constraints are real. I’m open to focused asynchronous projects and long-term embedded opportunities with international teams.
[email protected]