- Heading and scaling the Machine Learning team- Pushing the state-of-the-art on mixed Arabic/English financial document understanding- Work on methods for optimizing our internal Annotation Team

- Teaching Deep Learning and Neural Networks- Teaching my students how to demystify Deep Learning so they can read papers, understand where methods come from and make them usable later on in a production environment- Syllabus : Basics (MLPs), Computer Vision (CNNs, Diffusion Models (DDPM)), Natural Language Processing (RNNs, Transformers) and how to apply these architectures to real world applications.- You'll find the latest Practical Works for my students here: https://cpcdoy.github.io/articles/

• First employee as a Founding Machine Learning Engineer• The company went from 3 to 27 people in less than 2 years and raised a Series A round• Worked on setting up the initial pipeline for transaction categorization:• Named entity extraction (NER): Extraction of entities from (non-)natural language consumer andbusiness bank transactions in multiple languages using "cycles training" to avoid forgetting• Transaction categorization: Find the category of a transaction from a list that is not set and thatcan change based on what our customers want using a BERT encoder for zero-shot classificationapproach• Researched and improved NER using a Bi-GRU augmented tokenizer and a custom noise-robustmulti-head DeBERTa with a Knowledge Base (an approach I developed so the model can use newinformation without any retraining)• Built an in-house human labeling team with pay, testing, and training, as well as a hierarchy of labelersand reviewers• Labelers would receive a "cluster labeling task" which are similar tasks grouped together forbetter pattern clarity and greater labeling efficiency• Developed a fast and configurable generative model to help our sales team unblock our customersfrom benchmarking with us if they face data privacy issues or even lack any data• Worked on LLaMa-based models to future proof our pipeline. Also researched better LLMinterpretability using gradient-based methods• Presented at PyData 2022

- Work on a fast and accurate semantic similarity search engine using a state-of-the-art multilingual DistilBERT-based model and ANNG search in Python and Rust- Open-sourced my Rust implementation of our NLP model for real-time inference and was a major contributor of v0.3.0 of the sentence-transformers NLP library- Deployed our microservice architecture in production using Kubernetes and a complete stack of tools for benchmarking, testing, monitoring, logging, and reporting- Prototype work on a fast high-quality face swap/blend algorithm

- Work on DLSS (Deep Learning Super Sampling), one of Nvidia Turing's major feature for real-time anti-aliasing and upscaling on latest video games- Participated in conferences like GDC, GTC and SIGGRAPH- I also worked on other smaller projects related to deep learning like style transfer for portraits with autoencoders, and more

Development of a smart CCTV system to be used in data-centers.

- Development of an OS independent product trading platform for internal use, in PHP and JavaScript- Design a solution locally, document, test and then deploy the solution to the network- Work on creating a user-friendly and modern interface