01Implemented
Data FinOps
Specification-driven assessment and evidence-led reporting.
DATA · ML · AI PLATFORM
Systems that hold up when someone asks how they work.
I build data platforms and AI systems that teams can inspect, reproduce, and operate. More than twelve years across software, data engineering, machine learning, and cloud delivery.
SELECTED WORK
Public evidence, not a tool list.
- Implemented
- Code and documentation are public. This alone proves no runtime behavior.
- Locally validated
- A dated local command, test, or evidence record proves the stated behavior.
- Published dataset
- The Kaggle slug and version are recorded in the repository.
- 02Published dataset
Public data releases
Brazilian education and procurement data contracts for reproducible releases.
Two Kaggle datasets, version 1 each. PNCP procurement history: 1,979 rows per layer, notices from 2025-01-01 to 2025-01-07, modality 6. Education data lake: SIOPE annual municipal declarations 2019-2023 joined to IBGE codes, 27,830 municipality-year rows; a clean download was hash-verified.
- 03Locally validated
Education MLOps
Traceable municipality-level anomaly triage from public data.
v0.2.0 trains a robust peer-group anomaly-triage model on the pinned education Kaggle release. The 2022 batch was scored (96 review signals); the drift gate blocked the 2023 batch (spread ratio 1.384).
Outputs are review signals only. There are no labels, so no accuracy is claimed.
- 04Locally validated
Procurement ranking
Transparent retrieval and ranking with responsible matching limits.
v0.2.0 ranks historical notices from the pinned PNCP Kaggle release and verifies its hash before use.
Offline evaluation uses synthetic profiles with rule-derived judgments, so it measures constraint adherence, not user relevance.
- 05Implemented
RAG observability
Evaluated retrieval, source citations, and redacted traces.
- 06Locally validated
Distributed runtime
Redis-coordinated workers, Kubernetes, Terraform, and recovery tests.
Local Compose benchmark on 2026-09-24: 48 requests at concurrency 6, 61.89 req/s, p95 147.95 ms. Worker recovery proven on a local kind cluster.
Deterministic model stub on one Docker Desktop host; not a capacity claim. Cloud not validated.
CAPABILITIES
Each capability points to a public repository.
Data platforms and releases
Data contracts, reproducible public releases, and cost assessment.
ML systems
Traceable pipelines, offline evaluation, and transparent ranking.
GenAI and RAG
Retrieval evaluation, source citations, and redacted traces.
Distributed runtime and delivery
Redis coordination, Kubernetes, Terraform, and recovery tests.
EXPERIENCE
From software foundations to data and AI platforms.
2024 — present
Data and AI engineering
Data pipelines, AI applications, governance, and observability.
2023 — 2024
Data and machine learning
Lakehouse pipelines, MLflow monitoring, and versioned data assets.
2022 — 2023
GCP data and ML
Composer, Dataproc, Vertex AI, and CI/CD for data systems.
2014 — 2022
Software, data, and cloud engineering
Data platforms, full-stack systems, databases, and embedded software.
WORKING PRINCIPLES
Clear contracts. Measured behavior. Public boundaries.
I use reproducible data releases, explicit evaluation, infrastructure-as-code, and documented limits to make technical work reviewable.
CONTACT