Chady Dimachkie

Head of Machine Learning

2023 to 2026

Machine learning for SME lending and financial document understanding.

• Led Kafalah ML delivery. Took the machine learning work through development and delivery with the Abwab team for Kafalah, a Saudi government entity supporting SME credit guarantees.
• Developed Spatial Text Rendering. Preserved document layout for text only LLMs, including financial documents mixing Arabic and English.

Founding ML Engineer

2021 to 2023

Making bank transactions understandable across languages.

• Built the initial transaction enrichment pipeline. Combined multilingual named entity extraction with adaptable zero shot transaction categorization.
• Advanced multilingual NER. Developed a BiGRU augmented tokenizer, a noise robust multihead DeBERTa and a knowledge base approach that accepts new information without retraining.
• Built the human labeling operation. Established hiring, testing, training and a labeler and reviewer hierarchy, with clustered tasks for better pattern clarity.
• Unblocked privacy constrained benchmarks. Developed a fast configurable generative model for prospects unable to share or supply data.
• Explored the next generation pipeline. Worked on LLaMA based models and gradient based LLM interpretability.
• Presented at PyData 2022. Shared the team’s work with the data community. View the talk

Machine Learning Engineer

2019 to 2021

Production NLP for games and player communities.

• Built multilingual semantic search. Combined a DistilBERT based model with approximate nearest neighbour search in Python and Rust.
• Worked on real time automated moderation. Filtered toxic in game comments and usernames and researched character level Transformers.
• Open sourced production NLP work. Created rust-sbert and contributed to Sentence Transformers 0.3 and rust-bert.
• Deployed ML microservices in production. Used Kubernetes with benchmarking, testing, monitoring, logging and reporting.
• Prototyped high quality face swapping. Explored a fast approach alongside the production NLP work.

rust-sbert on GitHub127 stars

Credited on these games

Assassin’s Creed ValhallaMobyGames credits

Watch Dogs: LegionMobyGames credits

Hyper ScapeMy credit in the EMEA archive

My MobyGames credits

Deep Learning Intern

2018 to 2019

Deep learning for real time graphics.

• Contributed to the first generation of DLSS. Worked on NVIDIA’s AI anti aliasing and image upscaling technology for games.
• Integrated deep learning into Unreal Engine 4. Worked inside the engine while developing the early DLSS integration.
• Explored portrait style transfer. Prototyped autoencoder based image transformation.
• Taught at major graphics conferences. Developed a tutorial and assisted sessions at GDC, GTC and SIGGRAPH.