deyab
Abd Alrhman Deyab
Computer engineering student building applied machine learning and full-stack systems.
0%
Scroll
Abd Alrhman Deyab
Computer Engineer
Design, Full-Stack Development, and AI Enhancement
About
Welcome to my portfolio! You can call me Deyab! I’m from Brooklyn, NY, and I’m currently studying abroad for my engineering degree in Istanbul, Turkey at Istanbul Medipol University! I’m currently in my junior year of university. I speak English and Arabic fluently, intermediate Turkish, and currently learning Russian. I created this website to mainly showcase my resume and work, and to also showcase my attention for details and web development capabilities. My hobbies are Art, Music, Photography, Watching and Reviewing Movies, Fashion, Travelling, and Language Learning. I hope you find this portfolio appealing! :)
Projects
Study LLM
Notebook RAG chat
Document-grounded RAG platform: course materials organized into notebooks, answers cited to specific source chunks, and an explicit “not found in sources” response to suppress hallucination. Retrieval recall raised by widening each vector hit to adjacent chunks after measuring page-break failures.
RepoSpring Boot, React, PostgreSQL/pgvector, Ollama, Docker
Flower-VLM
Vision-language identification chat
Chat that answers flower queries from text or uploaded photos by embedding both into a shared CLIP space over 7,390 indexed images. Vector search runs before generation, so descriptions explain the matched species instead of guessing blind.
RepoNode.js/Express, React, Qdrant, CLIP, Ollama
Flower Semantic Search
Text and image search engine
About 7,370 Oxford-102 flower images indexed as L2-normalized CLIP embeddings, searchable by natural-language description or example image, with taxonomy-aware query expansion and a pluggable Qdrant or FAISS vector store.
RepoPython, FastAPI, React, CLIP, Qdrant/FAISS
Medical Articles Platform
Server-rendered social platform
A social platform for medical articles with session-based authentication, Django ORM models, and a normalized SQLite schema.
RepoPython, Django, SQLite
Study LLM
Notebook RAG chat
Document-grounded RAG platform: course materials organized into notebooks, answers cited to specific source chunks, and an explicit “not found in sources” response to suppress hallucination. Retrieval recall raised by widening each vector hit to adjacent chunks after measuring page-break failures.
RepoSpring Boot, React, PostgreSQL/pgvector, Ollama, Docker
Flower-VLM
Vision-language identification chat
Chat that answers flower queries from text or uploaded photos by embedding both into a shared CLIP space over 7,390 indexed images. Vector search runs before generation, so descriptions explain the matched species instead of guessing blind.
RepoNode.js/Express, React, Qdrant, CLIP, Ollama
Flower Semantic Search
Text and image search engine
About 7,370 Oxford-102 flower images indexed as L2-normalized CLIP embeddings, searchable by natural-language description or example image, with taxonomy-aware query expansion and a pluggable Qdrant or FAISS vector store.
RepoPython, FastAPI, React, CLIP, Qdrant/FAISS
Medical Articles Platform
Server-rendered social platform
A social platform for medical articles with session-based authentication, Django ORM models, and a normalized SQLite schema.
RepoPython, Django, SQLite
Study LLM
Notebook RAG chat
Document-grounded RAG platform: course materials organized into notebooks, answers cited to specific source chunks, and an explicit “not found in sources” response to suppress hallucination. Retrieval recall raised by widening each vector hit to adjacent chunks after measuring page-break failures.
RepoSpring Boot, React, PostgreSQL/pgvector, Ollama, Docker
Flower-VLM
Vision-language identification chat
Chat that answers flower queries from text or uploaded photos by embedding both into a shared CLIP space over 7,390 indexed images. Vector search runs before generation, so descriptions explain the matched species instead of guessing blind.
RepoNode.js/Express, React, Qdrant, CLIP, Ollama
Flower Semantic Search
Text and image search engine
About 7,370 Oxford-102 flower images indexed as L2-normalized CLIP embeddings, searchable by natural-language description or example image, with taxonomy-aware query expansion and a pluggable Qdrant or FAISS vector store.
RepoPython, FastAPI, React, CLIP, Qdrant/FAISS
Medical Articles Platform
Server-rendered social platform
A social platform for medical articles with session-based authentication, Django ORM models, and a normalized SQLite schema.
RepoPython, Django, SQLite
All repositories live on GitHub
Experience
ML Dev Intern, MAG7 Solutions
- Contributed end to end to a web app using machine learning and image-based product recognition: embedding pipeline, model integration, and API delivery.
- Designed, built, and deployed three full-stack AI systems covering multimodal semantic search, vision-language chat, and RAG over user-uploaded documents.
- Containerized multi-service stacks with Docker Compose and Docker Swarm: multi-stage builds, service networking, persistent volumes.
- Profiled and optimized local model inference, cutting cold model load about 7.5 times and scanned-document ingestion from about 11.5 minutes to about 23 seconds.
Resume
The full document as a PDF: education, experience, projects, and the complete skills list in one page.
Download Resume- Languages
- Python, Java, JavaScript, SQL, HTML/CSS, MATLAB
- Backend and frontend
- FastAPI, Spring Boot, Django, Node.js/Express, React, Vite, REST API design, JWT and bcrypt authentication
- Machine learning and AI
- CLIP, embeddings, semantic and vector search, RAG, LLM and VLM integration (Ollama), Hugging Face Transformers, OCR (Tesseract), chunking strategies, inference optimization
- Data and storage
- PostgreSQL/pgvector, Qdrant, FAISS, SQLite, SQLAlchemy
- DevOps and tools
- Docker, Docker Compose, Docker Swarm, Git/GitHub, Linux/SSH, Maven, JUnit, Testcontainers