Abwab.ai
Head of Machine LearningI lead machine learning at Abwab.ai. My recent work includes leading ML development for Kafalah, a Saudi government program that helps small and medium businesses get loans.
Chady Dimachkie
Head of Machine Learning at Abwab.ai.
Previously: founding ML engineer at Ntropy, ML engineer at Ubisoft, and contributor to the first generation of NVIDIA DLSS.
I also teach deep learning at EPITA. My open source contributions include Hugging Face projects, Guidance (originally Microsoft) and MySQL.
Experience
2023 to 2026
Ntropy
Founding ML EngineerHelped build the initial transaction enrichment pipeline: extracting names and businesses from bank transactions in multiple languages, then categorizing those transactions.
Ubisoft
Machine Learning EngineerWorked on semantic search and automated moderation of in game text, and deployed ML services in production. Created rust-sbert, a Rust port of Sentence Transformers, and contributed to Sentence Transformers and rust-bert.
NVIDIA
Deep Learning InternContributed to the first generation of DLSS, NVIDIA’s AI image upscaling technology for games, and its integration into Unreal Engine 4.
About NVIDIA DLSSOpen source contributions
Hugging Face
Triplet losses in Sentence Transformers; inference fixes in Lighteval.
Sentence TransformersLighteval
Microsoft / Guidance
GPU device handling and prompt debugging in Guidance, originally a Microsoft project.
Guidance contributions
Rust ecosystem
BERT and RoBERTa pooling in rust-bert; author of rust-sbert.
rust-bert contributionrust-sbertEPITA · MASTER’S STUDENTS · PARISAll course materials & tutorials
Teaching deep learning
From building a first neural network to implementing an image generation model. My course combines practical work in PyTorch with the habits needed to read and question current research.
01
Neural networks in PyTorch
Build and train networks, understand losses, and experiment with optimization.
Explore the practical02
Convolutional networks
Understand convolution, build CNNs and improve their performance on image tasks.
Explore the practical03
Diffusion & image generation
Implement a DDPM, from adding noise to a U-Net that learns to remove it.
Explore the practical04
Reading research critically
Identify assumptions, inspect experiments and decide whether a paper is useful.
Explore the practicalWriting
July 2023
Cross Entropy is All You Need… Or is It?
A loss function for noisy labels and better model calibration.
Read the article
May 2023
Efficiency and Maintainability in Named Entity Recognition: A Trie Based Knowledge Base Approach
Using an external knowledge base to improve and update an NER model.
Read the article
March 2022
Multilingual Named Entity Recognition for Financial Transactions
Training one transaction enrichment model across multiple languages.
Read the articlePersonal projects
Eclipse Times
My eclipse guides for 2026 and 2027: city timings, maps and a 3D viewing simulator, in English, French and Spanish.
Outside work
These days I’m into building speakers, making music, playing badminton, building a mechanical watch, 3D printing with a Bambu Lab P2S, and experimenting with ways to orchestrate AI agents.