FlowCV Logo
Education
Publications

  • •Achieved 95.85% accuracy with GB after model evaluation and hyperparameter tuning.

  • •Applied Explainable AI (XAI) techniques (LIME, SHAP) to identify important factors affecting predictions.
  • •This study was funded by the National Research Foundation of Korea (NRF) grant funded by the Korean government and the MSIT (Ministry of Science and ICT)
    • Professional Experience
    Outlier AI, AI Coder & Evaluation Specialist

    Contributed to the development and evaluation of advanced AI systems to support large language model (LLM) training. Focused on code generation, prompt refinement, and evaluation of model outputs using Python.

    Key Responsibilities & Achievements:

  • •Evaluated and refined AI-generated Python code across multiple technical domains.
  • •Contributed to prompt engineering and multi-turn instruction tuning for LLMs.
  • •Participated in vision-to-code translation tasks using multimodal model inputs.
  • •Built and debugged Python environments for code execution and testing.
  • •Annotated architectural images and structured outputs for reasoning tasks.
  • •Rated and ranked AI responses using rubrics to improve model alignment (RLHF).
  • •Worked across diverse internal projects such as:
  • •Vibrant Clownfish: Iterative code generation and response ranking
  • •Hyperion: Environment setup and augmentation for SWEAP problems
  • •Tauros: Vision-based code generation from UI/image input
  • •Oasis Secure: Reconstructing GitHub issues from pull requests
  • •Ambassador Cafe & Hopper Code: Evaluation of agentic tools and multi-turn code reasoning

    • Technologies & Skills: Python · Prompt Engineering · Code Evaluation · RLHF · Multimodal Models · Environment Building · Vision-to-Code Translation, Github.

    Skills
    Programming Languages & skills: C++ | Python | Object-Oriented Programming (OOP) | Data Structures & Algorithms (DSA)
    Artificial Intelligence: Machine Learning | Deep Learning | Convolutional Neural Networks (CNN) | Natural Language Processing (NLP) | Model Fine-tuning | computer vision | Generative AI | GANs | Transformers | Diffusion | RAGs | LLMs
    Tools & Frameworks: TensorFlow |  PyTorch | Scikit-learn | Numpy, Pandas | Matplotlib, Seaborn | SHAP, LIME | HuggingFace | Git, GitHub | linux environments | Diffusers (Huggingface) | LangChain | OpenAI API
    Projects
  • •Developed a comprehensive web-based diagnostic platform for Alzheimer’s detection, integrating clinical data analysis, MRI scan processing, and AI-assisted diagnosis.
  • •Designed a multi-model AI architecture: XGBoost for structured clinical data, ResNet50 CNN and SWIN Transformer for MRI analysis, and fine-tuned Gemini Pro for medical report generation and patient communication.
  • •Integrated Vision Transformers (ViT) for enhanced spatial pattern recognition in brain MRI classification using HuggingFace models.
  • •Leveraged Gemini Pro for both structured diagnostic reasoning and synthetic medical image generation from textual descriptions, enabling advanced AI-assisted decision-making and data augmentation.
  • •Deployed the full-stack solution on Streamlit with secure AWS-backed data management, including role-specific dashboards for doctors, patients, and administrators.
  • •Enabled advanced neuroimaging capabilities such as region-of-interest measurements, detailed scan visualization, and longitudinal brain comparisons.

    • • Optimized healthcare workflows through EMR integration, automated patient registration, and smart scheduling tools.

    • Engineered prompt templates for real-time clinical consultation, making Gemini Pro act as an intelligent medical assistant.

    Real-Time Object Detection using ESP32-CAM
  • •Developed a real-time object detection model using the FOMO framework with MobileNetV2.
  • •Preprocessed a custom dataset of 97 images with data augmentation
  • •Achieved 94.44% accuracy and 90.9% F1 Score
  • •Optimized the model with quantization and EON™ Compiler
    • SageAI, Medical Chatbot RAG Application
  • •Developed a Retrieval-Augmented Generation (RAG) pipeline by integrating Faiss for dense vector search and Pegasus for generative medical response.
  • •Preprocessed a 637-page medical corpus and compared transformers (T5, Pegasus, BERT) for text embeddings and summarization—chose Pegasus for quality and efficiency.
  • •Used CUDA acceleration and parallel processing for scalable vector retrieval.
  • •Processed a dataset of 100,000+ books
  • •Used TF-IDF and cosine similarity
  • •Deployed the application using Gradio on Hugging Face Spaces
  • •Developed using ResNet50 for transfer learning.
  • •Preprocessed and augmented dataset 47,000+ Image
  • •Fine tuned ResNet50 using Adam optimizer and categorical cross-entropy.
  • •Used OpenCV for image reading and deployed the system using Streamlit
  • •Designed CNN architecture with convolutional, max-pooling, dropout, and dense layers using TensorFlow.
  • •Applied data augmentation for better generalization.
  • •Trained the model using Adam optimizer and categorical cross-entropy loss.
  • •Deployed using Tkinter for a graphical user interface with accuracy of 95%.
    • University Courses
    Structured Programming
    Artificial Intelligence
    Computer Vision
    Object Oriented Programming
    Machine Learning
    Knowledge based Systems
    Data Structures
    Natural Language Processing
    Optimization Techniques
    Algorithms
    Neural Networks

    Database Systems
    Awards
    ICPC ECPC Qualifications Collegiate Programming Contest Day 9, ICPC Egyptian Collegiate Programming Contest
    08/2022
    AWS Machine Learning Foundation, AWS