AI/ML engineer working on LLM evaluation, retrieval systems, and Arabic language resources. Previously a contractor on Meta's Generative AI Language Research team, evaluating frontier-model safety and conversational AI quality. M.Eng in Applied Artificial Intelligence, Stevens Institute of Technology.
Based in Providence, Rhode Island. I work in English and Arabic.
- Arabic WordNet 4.0: an Arabic wordnet with full coverage of Open English WordNet 2024. 120,630 synsets, 297,150 semantic relations, 184,238 senses, every synset carrying an Arabic definition. WN-LMF 1.4, CC BY 4.0, archived on Zenodo.
- wn-editor-extended: the editing library behind it, published on PyPI. Full CRUD over WN-LMF, automatic inverse relations, compound operations, and a 22-rule validation engine.
- Sal-Parsing: a benchmark of twelve PDF-to-Markdown parsers on scientific papers, including the Arabic OCR tier, with setup guides and side-by-side quality reports.
- PromptTree: a local Streamlit app that keeps prompts and model responses as branching conversation trees.
LLM evaluation and guardrails, retrieval over large knowledge bases, Arabic NLP, and the general problem of getting model output to a state where a system can act on it safely.
- Blog: salah-sal.github.io
- LinkedIn: in/salah-abdo-ms
- Email: salah.abdo.tech@gmail.com



