PROJECTS.17
A collection of my work in Machine Learning, Computer Vision, and Data Science.
CLI + TUI for Bare-Metal AI Infrastructure
Built a read-only Go CLI + TUI that inspects a single AI box — host, GPUs, Docker, and local LLM runtimes (Ollama, vLLM, SGLang, Triton) — and correlates them into one snapshot, rendered as CLI tables, a live terminal dashboard, JSON, or six MCP tools for coding agents. Cross-platform, packaged via Homebrew, APT, and a no-sudo install script.
MCP Server for AI Agents
Built an MCP server that gives any AI agent a styled SVG map of Jordan — the whole country or a single governorate — plus a bilingual directory of its 122 hospitals. Five typed tools over SSE, no API key and no network at call time: boundaries ship inside the package, so every render is offline and deterministic. Runs as a Docker Space on Hugging Face.
Prometheus Exporter
Built a Prometheus exporter for Open WebUI that polls its REST API — no database access required — and reconstructs per-model cost from an overridable price table. Ships with a 9-row Grafana dashboard covering usage, quality, and spend.
AI Interpretability Tool
Built an interactive tool that runs a live PyTorch forward pass through Qwen3-1.7B and streams attention, logit-lens predictions, and token probabilities to the browser in real time. No pre-baked demos — every visual comes from the user's own prompt.
Prometheus Exporter
Built a Prometheus exporter for self-hosted Qdrant that surfaces 34 per-collection, per-shard, and cluster metrics the native /metrics endpoint doesn't expose. Ships with an auto-provisioned Grafana dashboard and Docker Hub/GHCR images.
AutoML Platform
Designed and built a production-grade no-code AutoML platform with model training, evaluation, and deployment. ClickML is a powerful and user-friendly platform designed to empower non-technologists to create predictive models for their businesses effortlessly. Used by 5,000+ users, enabling end-to-end machine learning workflows without coding.
LLM Tooling – Python Package & Platform
Developed an open-source Python package (9K+ downloads) for structured prompt engineering and project context aggregation. Extended into a containerized, no-code platform deployed via Docker.
Computer Vision / Healthcare
Developed OneLook (formerly RetinAI), a computer vision platform that analyzes retinal images, automates disease diagnosis, and generates reports in under a minute. Awarded 2nd place at the 8th Hakeem Academy Competition.
GenAI Healthcare Platform
Implemented an LLM-powered platform using RAG (Retrieval-Augmented Generation) to help users understand medical prescriptions, including document ingestion, semantic search, and conversational interfaces for accurate information retrieval.
Python Package
Created an open-source Python package for Streamlit components with 8K+ downloads enabling correct Arabic text alignment and rendering.
Real-Time Computer Vision
Built a real-time deep learning system for face detection, tracking, and emotion recognition using YOLO-based object detection and multi-stage vision pipelines.
NLP
Built a full-stack Arabic search engine from 500 articles, including text preprocessing, indexing, ranking, and a web-based user interface using TF-IDF and cosine similarity.
A computer vision project with two phases: Playing Cards Detection and building a Tarneeb game based on model detections.
Web-based application to detect and blur all faces in an image using YOLO v8 and Streamlit.
Detect whether a person is wearing a face mask using Transfer Learning (ResNet50) and custom TensorFlow models.
Blur faces in images while keeping the rest unchanged using Haar Cascade and YOLO v8.
Art Styleify is a web application that allows you to transform your photos into stunning masterpieces using various artistic styles. Apply styles of famous artists like Van Gogh, Josef Sima, and more.