Skip to content
Free assessment

About

Osney A. de Souza, AI engineer

An AI engineer in Joinville, Brazil, with five years of software development and AI systems running in production. The person who handles your project is the one who designs it and writes the code.

01What he does today

He builds and runs an LLM-assisted audit platform in production. It reads unstructured documents (scanned PDFs, Excel, bank files, CSV), answers with cited sources (RAG with embeddings and pgvector) and uses agents with tool calling, MCP and structured outputs.

Local and cloud models work in a cost-based cascade, and a confidentiality gate decides what may go to the cloud. Thousands of automated tests, CI/CD, and monitoring with Prometheus and Grafana keep it running.

02The central rule

The model never produces the final number. Values come from reading the document or from recalculation in code, and a deterministic checker vetoes wrong mappings.

The same rule applies to every project on this site: AI speeds up reading and interpretation; the checking is done by code that can be audited.

03Before that

  • Computer vision at scale. License plate recognition (LPR) with convolutional neural networks for video monitoring handling petabytes of video per day, plus semantic search over the metadata.
  • Systems and apps. Back ends in Node/Express, NestJS, .NET and Java/Quarkus; apps in Flutter, React Native, Kotlin and Swift; infrastructure on AWS EKS and GitHub Actions.
  • Industrial automation. Python bots, SAP HANA and integration between ERP and the shop floor.
  • Remote work for a US company. Payments with Stripe, messaging with Twilio and observability with OpenTelemetry.

04Projects and education

Climate-risk analysis project for ESG, with PostGIS, scikit-learn and SHAP to explain each prediction.

Studying Software Engineering at Univille, expected to graduate in 2027.

Everyday tools

  • Python
  • PostgreSQL and pgvector
  • Ollama and local models
  • OpenAI and Anthropic
  • OCR and computer vision
  • FastAPI
  • Flutter
  • Docker and Kubernetes
  • Prometheus and Grafana

05How projects are run

Pilot before production

A small scope, a short timeline and metrics agreed before we start. The decision to continue is yours, with the numbers on the table.

Every number has a source

Every value that enters your system has an origin: a cell, a page, a line or a calculation. Anything that doesn't reconcile goes to review.

Confidentiality and privacy

A confidentiality agreement, minimal access and, when the data is sensitive, models running on your own servers.

No conflicts of interest

Projects that could conflict with existing professional commitments are declined at the assessment stage, openly.

Open-source work on GitHub

Contact

Let's talk

Projects, partnerships or a technical question: write with some context and you'll get a straight answer.

Send an emailjuniorthesouza017@gmail.com Message on WhatsApp(47) 98864-2296

Tell us about the process, the rough volume and the systems involved. You'll hear back from the engineer who would build it.