Abdellatif Dalab, Developer in Montreal, QC, Canada
Abdellatif is available for hire
Hire Abdellatif

Abdellatif Dalab

Verified Expert  in Engineering

Data Scientist and Software Developer

Location
Montreal, QC, Canada
Toptal Member Since
December 8, 2021

Abdellatif is a machine learning engineer and data scientist passionate about mathematics and applied research. He currently leads the ML engineering initiative at Repustate, where he develops and deploys custom, multilingual deep learning models for various natural language applications. Abdellatif also designs and engineers deep learning API servers to support the company's core natural language services.

Portfolio

Intramove.ai
Amazon EC2, Docker, PyTorch, PyPI, Stripe API, MongoDB, FastAPI, Python, GitHub...
Repustate
Artificial Intelligence (AI), Amazon S3 (AWS S3), Algorithms, Deep Learning...
Decathlon
Python, SQL, Jenkins, Keras, TensorFlow, PyTorch, Scikit-learn, Pandas, NumPy...

Experience

Availability

Part-time

Preferred Environment

Google Colaboratory (Colab), Visual Studio Code (VS Code), Jupyter Notebook, MacOS, GitHub

The most amazing...

...project I've worked on at Repustate is designing and developing a new generation v5 DL microservices API that communicates with a monolith Go-based application.

Work Experience

Founding Engineer

2022 - PRESENT
Intramove.ai
  • Engineered core microservices to process payments, register users, keep track of API credits, and perform text analysis.
  • Created a PyPI API client for developers to interact with the server.
  • Built a website to demonstrate Intramove.ai, an API product for developers interested in analyzing economic news. The core service is powered by SOTA deep learning models for natural language and has endpoints to analyze headlines and articles.
Technologies: Amazon EC2, Docker, PyTorch, PyPI, Stripe API, MongoDB, FastAPI, Python, GitHub, Amazon Route 53

Senior Machine Learning Engineer

2022 - PRESENT
Repustate
  • Developed interpretable, supervised deep learning solutions for text using PyTorch and for implementing and improving research paper algorithms, such as BERT, custom attention layers, and hierarchical attention networks.
  • Created interpretable, unsupervised deep learning algorithms that eliminated Repustate's need for tagged data when building custom models, enabling scalable growth and taking the company's core service to the next level.
  • Designed and developed a new generation gRPC microservices API, allowing the main application developed in Go to communicate with Python deep learning servers.
  • Increased inference speed by 2-3x on average for prediction tasks using different techniques, such as ONNX quantization or serving and chunks batching.
  • Reduced the turnaround time for onboarding clients with custom needs from an average of 4 weeks to 1-3 days using the new generation API and ML pipeline.
  • Built an in-house multilingual transcription service that replaced Amazon Transcribe, reducing annual expenses by 14x.
  • Managed a multi-client portfolio, led technical discussions on sales calls, and aided in landing 10+ clients using the new generation API.
  • Devised and developed Amazon S3 schemas for production models and tokenizers.
  • Delivered custom deep-learning Docker images and RPM packages to be installed in on-premise servers.
  • Architected and developed a custom MLflow tracking server to record and monitor experimentation results and artifacts.
Technologies: Artificial Intelligence (AI), Amazon S3 (AWS S3), Algorithms, Deep Learning, Open Neural Network Exchange (ONNX), Hugging Face, gRPC, Protobuf, Git, Go, Python, PyTorch, Keras, MLflow, Amazon EC2, Docker, NumPy, Scikit-learn, Pandas

Data Scientist

2020 - PRESENT
Decathlon
  • Developed an in-house data-visualization pipeline that replaced a licensed tool saving $60,000 per year. Used SQL, Git, Jenkins, AWS cloud, Google Sheets API, and Google Data Studio.
  • Prototyped an NLU solution with customer reviews classification, keyword extraction, and sentiment analysis that outperformed a licensed tool, saving the marketing team $15,000 per year.
  • Created a visual search engine that was deployed as a product retrieval API. It's currently being used for product recommendations.
  • Built an unsupervised topic modeling solution for customer reviews with visualization, using sentence transformers. Improved original solution using GPT2 and prompt engineering.
  • Developed a store turnover forecasting tool using additive models and custom-made regressors (Prophet API).
  • Engineered an NLU product-article recommendation solution as part of Decathlon's personalization strategy.
  • Worked on data extraction, transformation, and loading tasks for each solution.
  • Made a sustainability reporting tool to monitor the performance of second-life and eco-designed products.
  • Built a color detection solution using k-means clustering to aid internal object detection models.
  • Interviewed new data science candidates, actively contributed to the hiring process, and mentored new interns on various data-related tasks.
Technologies: Python, SQL, Jenkins, Keras, TensorFlow, PyTorch, Scikit-learn, Pandas, NumPy, Flask, Google Data Studio, Redshift, Amazon S3 (AWS S3), Google Cloud Platform (GCP), GitHub

Machine Learning Intern

2019 - 2019
Decathlon
  • Developed and deployed a deep recommendation model for user-click item prediction using LSTM RNN architecture.
  • Surpassed the benchmarked precision, recall, and coverage metrics by improving the solution using attention models.
  • Developed and deployed an object detection model using TensorFlow's API for a hockey-brand detection application.
Technologies: Python, Algorithms, Keras, SQL, TensorFlow, GitHub

Machine Learning Developer Intern

2018 - 2018
Societe Generale
  • Developed a BI reporting tool using MicroStrategy.
  • Contributed to data visualization projects using Tableau.
  • Helped develop a web application using the Django framework.
Technologies: Tableau, Python, Django, MicroStrategy

AI Game Project and Report

https://github.com/abdelatifsd/Comp-472---MiniMax-Algorithm
This project was part of the AI course (COMP 472) at Concordia University. It includes a detailed report of the implementation strategy, minimax algorithm, and outcome analysis. The Jupyter Notebook has most of the code. The project didn't utilize any existing libraries. All algorithms and heuristics were implemented from scratch.
2015 - 2019

Bachelor's Degree in Information Technology

Concordia University - Montreal, Quebec, Canada

SEPTEMBER 2020 - PRESENT

Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning

Coursera

JANUARY 2018 - PRESENT

Improving Deep Neural Networks: Hyperparameter Tuning, Regularization, and Optimization

Coursera

JANUARY 2018 - PRESENT

Structuring Machine Learning Projects

Coursera

AUGUST 2017 - PRESENT

Neural Networks and Deep Learning

Coursera

JULY 2017 - PRESENT

Machine Learning with Python in Data Science

Udemy

APRIL 2017 - PRESENT

Machine Learning

Stanford University | via Coursera

Libraries/APIs

Keras, Scikit-learn, NumPy, Pandas, PyTorch, TensorFlow, React, Protobuf, Stripe API

Tools

Jenkins, GitHub, Tableau, Open Neural Network Exchange (ONNX), Git, PyPI

Languages

Python, SQL, HTML, CSS, JavaScript, Go

Storage

Redshift, PostgreSQL, Amazon S3 (AWS S3), MySQL, Google Cloud Datastore, Database Programming, MongoDB

Platforms

Google Cloud Platform (GCP), Visual Studio Code (VS Code), Jupyter Notebook, Docker, MacOS, Amazon EC2

Frameworks

Flask, Django, gRPC

Paradigms

Database Design

Other

Artificial Intelligence (AI), Data Analysis, Machine Learning, Natural Language Processing (NLP), Generative Pre-trained Transformers (GPT), Google Data Studio, Deep Learning, OOP Designs, Data Structures, Algorithms, Programming Languages, Hugging Face, MicroStrategy, Google Colaboratory (Colab), MLflow, FastAPI, Amazon Route 53

Collaboration That Works

How to Work with Toptal

Toptal matches you directly with global industry experts from our network in hours—not weeks or months.

1

Share your needs

Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2

Choose your talent

Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3

Start your risk-free talent trial

Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.

Top talent is in high demand.

Start hiring