85%
Less manual classification work at TRF1
Judicial analyst workflow improved through a production NLP delivery.
Senior AI Engineer / Software Engineer
Jonathan Oliveira
Senior AI Engineer | Software Engineer
I build AI systems that stay useful after integration, rollout, and operator feedback.
My work sits where model quality meets APIs, workflows, and institutional constraints. I have shipped applied AI programs across courts and public-sector teams, with a focus on delivery that is measurable, maintainable, and trusted by the people using it.
Credibility
Programs and institutions where I have shipped applied AI work.
- TRF1
- CNJ / PNUD
- PGDF
- TST
- AI.Lab UnB
Impact
Concrete outcomes anchored in real institutional work.
Weeks to days
Retraining cycles shortened
Lifecycle improvements made updates faster and safer to release.
30+
Precedent categories surfaced
Topic modeling and semantic similarity work supported national precedent discovery.
National-scale
Institutional programs delivered
Work spanning TRF1, CNJ/PNUD, PGDF, TST, and AI.Lab.
Featured projects
Named initiatives with clear institutional context and measurable work.
Initiative Legal TechPublic Sector AI
CNJ / PNUD
PEDRO Precedent Discovery Platform
Data Scientist · Jul 2022 - May 2023
National-scale precedent discovery initiative for CNJ and PNUD, combining FastAPI services, unsupervised NLP, semantic grouping, and governed experimentation to systematize qualified precedents from Brazil's highest courts.
Impact
Enabled discovery of more than 30 precedent categories across long-form judicial decisions.
- FastAPI
- NLP
- Semantic Similarity
- MLflow
- Legal Tech
Key outcomes
- More than 30 precedent categories identified through semantic workflows
- AI services integrated with CNJ systems through REST APIs
Initiative Legal TechPublic Sector AI
PGDF
OSIRIS Legal-Fiscal AI Workflows
Data Scientist · May 2023 - May 2024
AI delivery for PGDF legal-fiscal operations, spanning production APIs, supervised and semi-supervised models, active learning, and early LLM exploration for document-heavy institutional workflows.
Impact
Brought governed ML workflows and production APIs into legal-fiscal operations, while designing active-learning paths for longer-term model adaptation.
- FastAPI
- Active Learning
- MLflow
- DVC
- LLM
Key outcomes
- Production APIs connected model outputs to PGDF internal systems
- Active-learning loop designed to reduce model drift over time
Experience
Delivery history that explains the outcomes.
The strongest signal in my profile is a pattern of shipping AI work inside institutions where adoption, explainability, and operational fit matter.
TTY2000 | TRF1
Jan 2024 - Present
Senior Machine Learning Engineer
Brasília, Brazil
Senior machine-learning engineer leading modernization of judicial NLP services at TRF1, spanning legacy-model refactors, orchestration, APIs, retraining flows, and analyst-facing delivery in a large federal court environment.
- Refactored legacy ML systems and reduced processing time by 25%.
- Delivered analyst-facing Django APIs consumed by more than 500 internal users.
- Automated ETL, retraining, and release flows with Airflow and CI/CD.
- Python
- Django
- PostgreSQL
- Airflow
- DVC
AI.Lab UnB
Mar 2021 - May 2024
Machine Learning Engineer
Brasília, Brazil
Machine learning engineer at AI.Lab UnB delivering legal-tech R and D and production ML systems across CNJ/PNUD, PGDF, TST, and TRF1 initiatives, connecting stakeholder discovery, NLP, APIs, MLOps, and technical leadership.
- Contributed to PEDRO, OSIRIS, SABIA, and ALEI across legal and institutional programs.
- Built supervised, unsupervised, and semi-supervised models for production systems.
- Standardized MLOps patterns for experiment tracking, dataset versioning, and repeatable delivery.
- Python
- FastAPI
- Flask
- MLflow
- DVC
Arvvo Tecnologia
Jan 2019 - Mar 2020
Software Engineer Intern
Brasília, Brazil
Started in full-stack delivery, building internal products with Node.js, Vue.js, and MongoDB in a consulting environment that shaped my backend, API, and team-delivery fundamentals.
- Built responsive internal interfaces with Vue.js and Angular.
- Implemented REST services in Node.js and Express.
- Worked in Git-based review workflows across frontend and backend.
- Node.js
- Vue.js
- Angular
- MongoDB
- Express
Skills
Technical range shaped by product and operational pressure.
AI / LLM
- RAG
- Multi-agent systems
- Prompt engineering
- Fine-tuning
- Knowledge graphs
- Context caching
- LangChain
- LangGraph
Backend
- FastAPI
- Flask
- Django
- REST API design
- Nginx
- Python
- Node.js
Data / MLOps
- MLflow
- DVC
- BentoML
- Airflow
- Docker
- GitLab CI/CD
- PostgreSQL
- MySQL
- Oracle
- SQL Server
Technical writing
Writing that supports the project work rather than replacing it.
Technical writing
Production RAG Systems Need More Than Retrieval Demos
Oct 13, 2025
A production RAG system should be treated as a retrieval and evaluation pipeline with explicit failure modes, not as a prompt wrapper around a vector store.
- RAG
- Evaluation
- Vector Search
- Production AI
Technical writing
LLM Evaluation in Production Starts With Explicit Failure Modes
Jul 2, 2025
Evaluation is most useful when it reflects the failures a system can actually produce in production: missing context, wrong retrieval, incorrect tool use, unstable outputs, and unhelpful responses.
- LLM
- Evaluation
- Production AI
- Quality
Technical writing
Scaling ML Pipelines Means Reducing Hidden Manual Work
May 19, 2025
ML pipelines usually fail to scale because they depend on undocumented manual steps around data preparation, retraining, packaging, and release coordination.
- MLOps
- Airflow
- MLflow
- CI/CD
Next step
Need someone who can connect applied AI ideas to reliable delivery?
If you are evaluating fit, start with the projects and the experience narrative. They show how I handle architecture, stakeholders, APIs, model lifecycle work, and delivery under institutional constraints.