I build AI systems for retrieval, evaluation, automation, and applied NLP.
2023 — Napolab
The Natural Portuguese Language Benchmark.
Portuguese language model evaluation benchmark and dataset collection. A key contribution of this work is FaQuAD-NLI, which has been widely reused by the Portuguese NLP community, including in Portuguese LLM evaluation tooling and leaderboards.
- Napolab Leaderboard
- Medium article: The Hidden Truth About LLM Performance
- Master’s thesis: Lessons learned from the evaluation of Portuguese language models
2021 — hashformers
State-of-the-art research code for multilingual hashtag and word segmentation.
Hashformers uses language models and beam search to segment hashtags and whitespace-free text. The project was recognized as state-of-the-art for hashtag segmentation at LREC 2022 and has been cited and reused in multiple research papers.
2022 — neuralmind-ai/coliee
Code for legal NLP research:
- To Tune or Not To Tune? Zero-shot Models for Legal Case Entailment
- Yes, BM25 is a Strong Baseline for Legal Case Retrieval
2020 — ruanchaves/BERT-WS
Code for:
2020 — ruanchaves/assin
Code for:
2020 — ruanchaves/elmo
Code for:
Selected contributions to machine learning and NLP libraries.
2022 — argilla-io/argilla
Fixed bugs and shipped features related to semi-supervised learning during my internship at Argilla.
2021 — huggingface/transformers
Modified the Trainer class to support simultaneous Ray Tune and Weights & Biases execution.
2021 — awslabs/mlm-scoring
Improved installation instructions for the mlm-scoring library.
2020 — facebookresearch/BLINK
Fixed a parameter bug in a BLINK benchmark script.
Fixed a severe bug in the evaluation procedure.
The fix was documented in the paper “Portuguese language models and word embeddings: evaluating on semantic similarity tasks”.
- Website: ruanchaves.github.io
- Email: ruanchaves93@gmail.com