Ahmad I. Elawady, Developer in Sheikh Zayed City, Giza Governorate, Egypt
Ahmad is available for hire
Hire Ahmad

Ahmad I. Elawady

Verified Expert  in Engineering

Bio

Ahmad is a machine learning researcher and engineer with a passion for building solutions from scratch. He is interested in formulating the thinking process of SMEs as machine learning solutions. During his two years of work experience, Ahmad developed machine learning solutions for different sectors, including a document layout extraction and image inpainting solution and a sophisticated system to find chemical synthesis plans.

Portfolio

Beyond Limits
AI Research, Natural Language Processing (NLP), Large Language Models (LLMs)...
Information Technology Institute
VMware vSphere, NVIDIA vGPU, VMware ESXi, IT Consulting, Mentorship & Coaching...
Integrant
Azure, Flask, Software Engineering, Python, Artificial Intelligence (AI)...

Experience

  • Machine Learning - 3 years
  • Python - 3 years
  • Research - 2 years
  • PyTorch - 2 years
  • Computer Vision - 2 years
  • Deep Learning - 2 years
  • Large Language Models (LLMs) - 2 years
  • Reinforcement Learning - 1 year

Availability

Part-time

Preferred Environment

Amazon Web Services (AWS), Python, PIP, Docker, PyTorch, Linux

The most amazing...

...task I’ve done is refactoring a complex C++ SDK, relying on a deep understanding of the business but little C++ skills, then guiding the development team.

Work Experience

Senior Unstructured Data Scientist

2023 - PRESENT
Beyond Limits
  • Designed and implemented a research tool to easily build and experiment with LLM-based applications through configurations that unified the research process, increased the experimentation throughput, and reduced the redundancy in the implementation.
  • Built a RAG system with open-source models that supports question answering, follow-ups, and small talk, enabling seamless, natural conversations. It serves as the foundation for our enterprise solution, which is used by clients with strict data privacy policies.
  • Conducted technical feasibility analysis for AI projects with the clients and helped scope feasible projects.
  • Enhanced the design of the team’s data preparation tool for training LLMs, resulting in reduced development time, improved code reusability, and easier extensibility.
Technologies: AI Research, Natural Language Processing (NLP), Large Language Models (LLMs), Large Language Model Operations (LLMOps), Open-source LLMs, Retrieval-augmented Generation (RAG), Image Retrieval, Text Retrieval, LangChain, LoRa, PyTorch, Developing AI Models locally, Machine Learning Operations (MLOps), model quantization, AI Chatbots, Chatbots, APIs, Fine-tuning, AI Agents, Vector Databases, Distributed Cloud, Cloud, DeepSpeed, Fully Sharded Data Parallelism (FSDP), Hugging Face, GRPO, LLM as a judge, LLM Evaluation BLEU - ROUGE, LLM inference, LM Evaluation Harness, Multi GPU training, Supervised Fine-tuning Trainer, Synthetic Data Generation, Text Generation Inference, DeepSeek, Datasets, Generative Artificial Intelligence (GenAI), Llama, OpenAI, Amazon OpenSearch, REST APIs, Text Classification, LlamaIndex, FastAPI, Supervised Learning, Small Language Models (SLMs), Speech to Text, Text to Speech (TTS), Unsupervised Learning, Reinforcement Learning from Human Feedback (RLHF), Data Extraction, Prompt Engineering, LangGraph, OpenAI API

Department Supervisor

2020 - 2024
Information Technology Institute
  • Provided technical consultancy on AI-related projects involving multiple cloud providers.
  • Oversaw the design and installation of the AI infrastructure to facilitate computational resource sharing among the system's users. It involved virtualization and networking.
  • Assisted in the curriculum planning and teaching efforts on the machine learning track—AI-Pro.
  • Mentored students in machine learning graduation projects.
Technologies: VMware vSphere, NVIDIA vGPU, VMware ESXi, IT Consulting, Mentorship & Coaching, Artificial Intelligence (AI), Pandas, NumPy, Artificial Neural Networks (ANN), Neural Networks, Motion Capture, Distributed Cloud, Datasets

Machine Learning Engineer

2021 - 2023
Integrant
  • Researched and developed solutions to do retrosynthesis planning. It was designed to find a sequence of reactions to synthesize a specific molecule from the available starting materials, such as an inventory.
  • Engineered the POCs so the client could demonstrate their ideas and evaluate the systems quickly.
  • Developed the API for the tools and deployed it on the cloud.
Technologies: Azure, Flask, Software Engineering, Python, Artificial Intelligence (AI), Pandas, NumPy, Artificial Neural Networks (ANN), Neural Networks, Developing AI Models locally, Machine Learning Operations (MLOps), APIs, Fine-tuning, Cloud, Supervised Fine-tuning Trainer, Datasets, REST APIs, FastAPI, Supervised Learning, Unsupervised Learning

Machine Learning Researcher

2019 - 2020
RDI
  • Researched and developed solutions to address problems related to the document's layout extraction, such as document orientation detection, tables extraction, and image inpainting using deep learning and classical computer vision techniques.
  • Engineered the POCs to demonstrate the ideas and evaluate the systems quickly. The POCs were used as a reference implementation to guide the development team's work.
  • Implemented a configurable ready-to-deploy pipeline for the OCR system that supports parallel calls to the microservices, handles call dependencies, and dynamically manages the optional calls.
  • Designed and implemented a set of tools to validate APIs requests and log the time each pre-specified function takes in each API call.
  • Developed custom tools for data processing, such as annotation and visualization.
Technologies: Optical Character Recognition (OCR), Computer Vision, Docker, Docker Compose, Flask, uWSGI, Object Detection, Artificial Intelligence (AI), Generative Adversarial Networks (GANs), OpenCV, Pandas, NumPy, Computer Vision Algorithms, Artificial Neural Networks (ANN), Neural Networks, Developing AI Models locally, Machine Learning Operations (MLOps), APIs, Fine-tuning, Supervised Fine-tuning Trainer, Synthetic Data Generation, Datasets, REST APIs, FastAPI, Supervised Learning, Speech to Text, Text to Speech (TTS), Unsupervised Learning, Data Extraction

Deep Learning Research Intern

2019 - 2019
Valeo
  • Conducted research in the fields of domain translation, sensor modeling, and video inpainting.
  • Developed computer vision algorithms to weakly annotate data.
  • Implemented a custom annotation tool to easily modify the segmentation annotation.
Technologies: Python, PyTorch, Deep Learning, Computer Vision, Image Annotation, Video Processing, Artificial Intelligence (AI), Generative Adversarial Networks (GANs), OpenCV, Pandas, NumPy, Computer Vision Algorithms, Artificial Neural Networks (ANN), Neural Networks, Developing AI Models locally, Fine-tuning, Supervised Fine-tuning Trainer, Synthetic Data Generation, Datasets, Supervised Learning

Experience

ReLIC: A Recipe for 64k steps In-context Reinforcement Learning for Embodied AI

https://arxiv.org/abs/2410.02751
A research project focused on training long-context policies using online reinforcement learning (RL), where I served as the lead author. I managed task distribution and tracked the team's progress. I led experiments to scale context length in online RL training with a transformer model to 64k steps, eight times longer than previous work. I adapted techniques from LLM training to efficiently store long-context visual data in memory and optimize model training.

Beyond Search | Hybrid Gen AI Solution

https://www.beyond.ai/enterprise-ai/gen-ai
A hybrid Gen AI solution that generates insights, automates tasks, and enhances decision-making while maintaining strict data privacy and control. As the data scientist, I led the AI component from inception, through PoC development, to productization. I fine-tuned open-source LLMs for our application, researched improvements to the data retrieval component, and added small talk support with guardrails for seamless, natural conversations. Based on clients' feedback, I continuously improved the system. I collaborated with the back-end team to design and implement APIs and worked with DevOps to deploy the solution.

Sotoor

https://sotoor.ai/home
An all-in-one optical character recognition (OCR) software that converts scanned documents, in any language, into fully editable and searchable files.

I was the machine learning researcher responsible for layout extraction and document generation.

Using a mix of deep learning and classical computer vision techniques, I developed a system to extract information such as the lines, the tables, the document's orientation, and the document's background or inpainting.

I also developed a pipelining system that manages how these functionalities are applied. It supports parallel calls to the microservices, handles call dependencies, and dynamically manages the optional calls. I built and deployed a prototype that served these functionalities through RESTful APIs.

RSynth — A Retrosynthesis Planning Tool

A machine learning-based tool to help the chemists with retrosynthesis planning. The project was my end-to-end responsibility. With the support of the SME, I developed tools to filter the chemical reactions, designed and trained ML models, and packaged and deployed the solution to be tested by the client.

Motion Capture Project

https://github.com/ITI-Mechatronics-40/motion-project-interface
At the core of the exercise analysis system, this project captures the motion, identifies the person, detects the actions, and estimates the person's pose. I led the team, designed the system, and deployed it.

YOLOv3D

A research project aimed at using the YOLOv3 to predict the surrounding vehicles' position and orientation for RGB cameras. It was developed as part of my participation in the Peking University and Baidu's Autonomous Driving competition on Kaggle.

Siameser

https://github.com/aielawady/Siameser
A Python module to embed or facilitate training of the feature extractor. The model is trained to minimize triplet loss. It was developed as a part of my late participation in State Farm's Distracted Driver Detection competition on Kaggle.

Horaira

https://github.com/aielawady/horaira
A Python module with the tools developed for Kaggle competitions. It includes image processing (e.g., circle centering and a veins highlighter), a pipeline wrapper, and data augmentation. It was developed as a part of my participation in APTOS' 2019 Blindness Detection competition handled on Kaggle.

Education

2022 - 2024

Master's Degree in Computer Science

Georgia Institute of Technology - Atlanta, GA

2018 - 2019

Professional Degree in Mechatronics

Information Technology Institute - Egypt

2013 - 2018

Bachelor's Degree in Mechanical Engineering

Mansoura University - Mansoura, Egypt

Certifications

MARCH 2022 - PRESENT

University Ambassador Program

NVIDIA Deep Learning Institute (DLI)

JANUARY 2021 - JANUARY 2024

AWS Certified Machine Learning

Amazon Web Services

AUGUST 2019 - PRESENT

Mechatronics

Information Technology Institute (ITI)

AUGUST 2019 - PRESENT

Deep Learning Specialization

DeepLearning.AI | via Coursera

MAY 2019 - PRESENT

Machine Learning

Stanford University | via Coursera

Skills

Libraries/APIs

PyTorch, Hugging Face Transformers, REST APIs, Keras, TensorFlow, OpenCV, Pandas, NumPy, DeepSpeed, OpenAI API

Tools

Amazon SageMaker, DeepSeek, Amazon Elastic Block Store (EBS), uWSGI, Docker Compose, VMware vSphere, NGINX, Amazon OpenSearch

Languages

Python

Frameworks

LlamaIndex, LangGraph, Flask

Paradigms

Synthetic Data Generation, Text Retrieval

Platforms

Amazon Web Services (AWS), Docker, Amazon EC2, Ubuntu, Azure, Linux

Storage

Amazon S3 (AWS S3)

Other

Deep Learning, Computer Vision, Machine Learning, Artificial Intelligence (AI), Natural Language Processing (NLP), Large Language Models (LLMs), Retrieval-augmented Generation (RAG), LoRa, Developing AI Models locally, Fine-tuning, Multi GPU training, Supervised Fine-tuning Trainer, Llama, Text Classification, Supervised Learning, Unsupervised Learning, Cloud, Video Processing, Convolutional Neural Networks (CNNs), Object Detection, Image Processing, Computer Vision Algorithms, Artificial Neural Networks (ANN), Neural Networks, Reinforcement Learning, Open-source LLMs, Transformers, Machine Learning Operations (MLOps), Recurrent Neural Networks (RNNs), model quantization, AI Chatbots, Chatbots, Motion Capture, APIs, AI Agents, Vector Databases, Distributed Cloud, Fully Sharded Data Parallelism (FSDP), Hugging Face, LLM as a judge, LLM Evaluation BLEU - ROUGE, LLM inference, LM Evaluation Harness, Text Generation Inference, Datasets, Generative Artificial Intelligence (GenAI), OpenAI, FastAPI, Small Language Models (SLMs), Reinforcement Learning from Human Feedback (RLHF), Data Extraction, Prompt Engineering, Numerical Methods, Robotics, Software Engineering, Optical Character Recognition (OCR), Image Annotation, Software Deployment, Research, PIP, Mechatronics, Embedded Systems, Computational Fluid Dynamics (CFD), NVIDIA vGPU, VMware ESXi, IT Consulting, Mentorship & Coaching, Team Leadership, Generative Adversarial Networks (GANs), Programming, Training, ICT Training, Deep Reinforcement Learning, AI Research, Large Language Model Operations (LLMOps), Image Retrieval, LangChain, GRPO, Speech to Text, Text to Speech (TTS)

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