
Wassim Seifeddine
Verified Expert in Engineering
Machine Learning Developer
Wassim is a software engineer with 7+ years of experience, including 4+ years in machine learning. He worked with various clients, from startups to research institutes to multinational corporations. Wassim stands out from the crowd because he is thorough about building scalable solutions that are adaptive to business requirements.
Portfolio
Experience
Availability
Preferred Environment
PyTorch, PySpark, NumPy, Jupyter Notebook, Pandas, Amazon Web Services (AWS), GPT, Generative Pre-trained Transformers (GPT), Natural Language Processing (NLP), Computer Vision, Python, Artificial Intelligence (AI), GPU Computing
The most amazing...
...project I’ve built is a legal case classification model, a system that classifies case descriptions and highlights crucial text elements.
Work Experience
Machine Learning Engineer
Animaj
- Cultivated a concept extraction from text, audio, and video content to understand social media trends better. Concepts include emotions, activities, and sentiments on what's going on in the video.
- Developed a set of Apache Airflow Directed Acyclic Graphs (DAGs) to orchestrate data and machine learning (ML) pipelines.
- Created a monitoring system to have a bird's-eye view of the whole system.
- Finetuned LLMS (gptj, gpt-neox, llama) on new datasets using memory and time-efficient techniques like LoRA, DeepSpeed ZeRO, and PyTorch's FSDP.
- Generated 2D/3D images and videos using Stable Diffusion and other generative AI models finetuned on a custom dataset.
- Experienced (research and industry) on quantized models for more efficient training and inference.
- Deployed LLMs on decentralized networks with gRPC servers and mesh networks.
NLP Engineer
Manada Technology LLC
- Acted as the AI advisor for building personalized GPT models on custom datasets.
- Helped optimize data ingestion and cleaning pipelines.
- Assisted the development team in optimizing their development and experimentation process.
NLP/Python Developer
Northeastern University - College of Engineering
- Worked on a tool to implement aspect-based sentiment analysis on product reviews.
- Developed a mechanism to summarize reviews based on aspects.
- Finetuned ML models on custom datasets to improve performance.
AI/ML Expert
1Bstories
- Developed a system to generate realistic-looking avatars based on multiple user choices to use with TTS models.
- Built a scene generation pipeline that generates images based on a specific description and preference.
- Deployed MLflow to production to monitor all ongoing machine learning model inferences.
AI/ML Specialist
Rich Lemon Apps FZE LLC
- Trained the AI model to generate user avatars in a specific style while preserving facial features.
- Achieved significant improvements compared to the previous approach using Dreambooth, resulting in avatars that closely resemble the original faces.
- Utilized strong knowledge of AI/ML, Python, LoRA, diffusion models, stable diffusion, and image processing to accomplish project goals.
Machine Learning Engineer
Martian Learning Inc.
- Helped with debugging evaluation issues for custom deep learning model.
- Worked with a team of AI researchers to debug the performance of the model router.
- Run the model on a GPU cluster for training and evaluation.
Machine Learning Engineer
Odem Global Pty Ltd
- Finetuned LLMs to deploy to a decentralized blockchain.
- Developed a gRPC server to offload inference to remote servers.
- Improved the inference speed of cutting-edge LLMs to deploy on more resource-constrained servers.
AI Developer
CodeComplete, Inc
- Engaged in the optimization of a causal language model's acceleration.
- Explored novel concepts to enhance performance by leveraging state-of-the-art technologies in the field.
- Conducted in-depth analysis and experimentation to identify potential bottlenecks and develop innovative solutions for optimizing the performance of the language model.
- Stayed abreast of the latest advancements in the field of language modeling and explored their applicability to improve further the acceleration and overall efficiency of the causal language model.
OCR AI Developer
ACFT PERFO
- Implemented a solution using AWS Textract to extract structured data from PDF files.
- Developed an end-to-end pipeline for extracting and saving the data into a data lake.
- Provided alternative solutions and tools to improve the efficiency and accuracy of the data extraction process.
Senior AI | Tech | HR Consultant
Block Born LLC
- Advised on AI tools to generate creative suggestions for game content production based on schema and scale.
- Reviewed and provided feedback on ideas related to implementing an AI tool for creative content suggestions.
- Demonstrated deep expertise in AI and its applications for creative content generation.
Machine Learning Engineer
Advest Marketing, LLC
- Worked on optimizing a product ads algorithm for social media.
- Researched techniques for improving product placement in videos.
- Explored deep learning techniques for generative modeling for videos.
ML Engineer with Skills in GPT-2/3
Toptal
- Trained a GPT-style model for simple language modeling to be deployed on a bit tensor network.
- Configured an iterative procedure to train models from previous checkpoints on new datasets.
- Monitored the model performance during and after training to know when to retrain.
Lead Machine Learning Engineer
Quantum Analytica
- Worked as a startup consultant for a real-estate startup to help them better understand their data infrastructure and guide them to the right tools using ETL pipelines, data lakes, delta tables, and hot storage.
- Developed a complete set of PySpark ETLs for transforming, cleaning, and normalizing data from different data sources and industries, including real estate and agriculture.
- Built a personalized ML-powered employee reward model as an MVP for an early-stage startup using customer-level data and rewards data from different providers.
- Managed the tech team, including designing the architecture for the whole team. The architecture ranges from web scrapers with dynamic proxies until the data is in hot storage, ready to be used by REST APIs.
- Developed machine learning models for consumer demand forecasters in the retail domain focusing on optimizing distribution to avoid out-of-stock.
- Worked on an algorithm for forecasting consumer demand for a certain set of products to help our client engage strategically in new markets.
PhD Researcher
Nantes Université
- Worked on my PhD thesis: The Acceleration of a Neural Network Using Computer Arithmetic.
- Devised new neural network training strategies to encourage better performances from low-precision neural networks.
- Developed custom CUDA operations for low-level operations and function approximations.
- Developed RL algorithm to solve optimization problems in large categorical observations spaces.
- Attended conferences and research schools related to machine learning and computer arithmetic.
Researcher
Inria
- Worked on a research project to enable running big machine learning models on small resource-constrained devices using early exit networks.
- Collaborated with two professors to do a literature review on model compression techniques, including quantization, pruning, and knowledge distillation.
- Implemented a reinforcement learning solution (DQN) to solve an optimization problem.
- Published a paper for an international conference and presented the work at the meeting.
Machine Learning Engineer
Navya
- Made an object detection model faster and lighter to be able to run in pseudo real time to be deployed on a self-driving car.
- Worked closely with the core machine learning team in order to make sure we were aligned on the experiment setup and results.
- Experimented with multiple model compression strategies like pruning, quantization, and compiling to evaluate the efficacy of each method.
- Compressed the model around 4x with 3x improvement in inference speed while maintaining the same performance as the original model.
Data Scientist
SannSyn
- Developed a system that classifies legal cases from raw text input by an industry expert. The data given was raw scraped and OCR'd pdf from online sources.
- Performed data analysis of unstructured data to understand what could be done with the data and the needed processes to improve the quality.
- Built a web scraper to scrape financial news.
- Provided a sentiment analysis feature for financial news articles using transformer-based models.
Machine Learning Engineer
Tedmob
- Developed a customer-facing chatbot using RASA AI, Dialogflow, and Microsoft bot framework for a leading telecom operator for handling FAQ and account-related questions with OTP authentication and third-party integrations.
- Tracked issues in real-time using various tools like Sentry, ELK stack, and Docker monitoring tools.
- Handled meeting with the client and gathered various team requirements for the best launch process.
Head of the iOS Department
Tedmob
- Led and oversaw a team of iOS developers for two years.
- Managed the hiring and onboarding process for new hires.
- Migrated the team tech stack and incorporated a new software architecture.
iOS Developer
Tedmob
- Developed commercial applications for clients ranging from one-person startups to multinational companies.
- Tracked issues happening on the application side in real time and resolved them in the next release.
- Migrated old applications from Objective-C to Swift.
Experience
Concept Extraction from Videos
The concepts include:
• Emotions
• Activities
• Sentiment
• Objects
• Movement
Quantized Neural Network for Object Detection
The model used is a MobileNetV2 pre-trained model on ImageNet with an SSDLite object detector. We trained the model in an FP32 data format.
I applied several model compression techniques to reduce the model's size and monitor its performance.
Some of the methods we used are:
• Quantization
• Pruning
• Fused convolution
• Knowledge distillation
The work concluded that the model could reliably detect objects in images with the same accuracy as the FP32 version while going as low as the INT8 data format.
Casual Language Model Fine-tuning
Legal Case Classification
• EDA on unstructured messy text data to understand what can be done with the data
• Build a model that would classify a legal case based on the description entered into several categories. This part aimed to provide a tool to help lawyers classify cases faster and more easily.
• Extract the entities from the case description relevant to the classification to help counterarguments.
Custom Language Model Training Framework
https://github.com/pegesund/nor_bertMulti-modal Text Classifier
Twitter Sentiment Analysis
Running Neural Networks on Edge Devices
https://ieeexplore.ieee.org/abstract/document/9664700Norwegian Sentiment Analysis Model
• Positive
• Neutral
• Negative.
The main difficulty was finding a good Norwegian labeled sentiment analysis dataset and fine-tuning an existing multilingual model.
Skills
Languages
Python, Python 3, Bash, SQL, Snowflake, C++, Swift
Libraries/APIs
PyTorch, NumPy, Pandas, Matplotlib, XGBoost, PySpark, TensorFlow, OpenCV, REST APIs, Scikit-learn, YouTube API, cuBLAS
Paradigms
Data Science, Mobile Development, ETL, App Development, Distributed Computing, Real-time Systems
Platforms
Jupyter Notebook, Docker, Databricks, Amazon Web Services (AWS), Embedded Linux, iOS, AWS Lambda, NVIDIA CUDA, Blockchain, Microsoft Edge, Google Cloud Platform (GCP)
Storage
NoSQL, MySQL, Amazon DynamoDB, Data Lakes, Amazon S3 (AWS S3)
Other
Machine Learning, Computer Vision, Sentiment Analysis, Deep Learning, Natural Language Processing (NLP), Artificial Intelligence (AI), Image Processing, OCR, Computer Vision Algorithms, Mobile App Development, Predictive Modeling, Algorithms, APIs, Data Analytics, Data Visualization, Forecasting, AI Design, Data Engineering, Deep Neural Networks, Data Analysis, CTO, Predictive Analytics, Statistics, Datasets, Causal Inference, Fine-tuning, Data Inference, Programming, Architecture, Analytics, Image Recognition, API Integration, GPT, Generative Pre-trained Transformers (GPT), Software Architecture, Graphics Processing Unit (GPU), Convolutional Neural Networks, Image Analysis, GPU Computing, Cloud Architecture, Models, Optimization, Consulting, Advisory, Technology Consulting, Startup Consulting, Facial Recognition, OpenAI, GPT-Neo, Google Publisher Tag (GPT), DeepSpeed, Text Generation, Speech Recognition, Web Development, Handwriting Recognition, Large Language Models (LLMs), Data Management, Stable Diffusion, ControlNet, Machine Learning Operations (MLOps), Data-driven Marketing, Generative Pre-trained Transformer 3 (GPT-3), OpenAI GPT-4 API, Research, Graph Theory, Arithmetic, Object Detection, Neural Networks, Slurm Workload Manager, Chatbots, Recommendation Systems, Quantization, Data Scraping, Clustering, Text Classification, Internet of Things (IoT), Language Models, Data Warehousing, Trend Forecasting, Topic Modeling, Revenue Projections, Video Analysis, Speech Synthesis, Sound, Text Animation, Diffusion Models, Image Generation, Quantitative Finance, Object Tracking, Generative Models, ChatGPT, Plugins, GPT-4, OpenAI GPT-3 API, Amazon Textract, YouTube Ads, Videos, LoRa, Flash Attention, Pricing Models, DreamBooth, Kohya, Wearables, Biometrics, Reinforcement Learning, Text to Image, Text to Video, MLflow, Midjourney, Avatars, DQN, Modeling, LLaMA, MosaicML
Frameworks
Apache Spark, Spark, Flask, gRPC
Tools
Amazon EKS, Xcode, Jira, Jupyter, Apache Airflow, AWS ELB, Amazon Elastic Container Service (Amazon ECS), AWS Fargate, Amazon SageMaker
Industry Expertise
Project Management
Education
PhD Degree in Computer Science
Nantes Université - Nantes, France
Master's Degree in Artificial Intelligence
ESIEE Paris - Paris, France
Bachelor's Degree in Computer Science
Lebanese University - Beirut, Lebanon
Certifications
Deep Learning Specialization
DeepLearning.ai | via Coursera
Advanced Machine Learning Specialization
DeepLearning.AI and Stanford Online | via Coursera
Machine Learning
Stanford University | via Coursera
iOS Developer Nanodegree
Udacity