AI/ML Engineer building applied machine-learning systems for agribusiness, one of the most data-rich and underserved verticals for applied AI globally. Work spans RAG APIs with hallucination detection, on-device computer vision, time-series forecasting, and end-to-end NLP pipelines, with an emphasis on tested, reproducible engineering (typed code, unit tests, CI, Docker).
Open to remote ML/AI engineering roles, international or Brazil-based.
| Project | What it solves | Highlight | Stack |
|---|---|---|---|
| sb100_agents | Producers lack scalable access to precise agronomic knowledge; agronomists are costly and research literature is inaccessible at field level | Self-hostable RAG API with a continuous 0.0 to 1.0 semantic-entropy hallucination score; 205 tests, ~83% coverage | Python · FastAPI · Qdrant · Ollama |
| visiosoil-app | Soil texture assessment requires lab analysis or trained specialists; neither viable for large properties in low-connectivity rural environments | On-device classifier across 5 texture classes; 3rd of 1,300+ at FETEPS 2025, paper accepted at ICPA/ConBAP 2026 | Flutter · Dart · TFLite · Riverpod |
| weather-forecasting | Short-term temperature forecasting for agriculture, energy and public-safety planning across 211 countries | LightGBM at 0.19°C RMSE, ~75% better than the Prophet baseline | Python · LightGBM · scikit-learn · Prophet |
| tweet-sentiment-analysis | Generic sentiment classifiers fail on social-media language; slang and platform-specific syntax cause unreliable outputs | RoBERTa fine-tuning pipeline with a 0.71 macro F1 zero-shot baseline; Rust CLI ~42x faster preprocessing | Python · Rust · HuggingFace · Polars |
Languages & Data
ML & Deep Learning
LLM & RAG
Backend & Mobile
Cloud & Tools
