Open to AI engineering opportunities

Abdullah AlJafari · AI Engineer

AI Engineer buildingdeployable ML systems.

I build computer vision, Arabic NLP/RAG, and MLOps pipelines from notebooks to tested APIs, Dockerized services, and cloud-ready deployments.

Signal 1

Computer Vision

Image inference

Signal 2

Arabic NLP/RAG

Retrieval + language

Signal 3

MLOps

API to deployment

Systems

Featured AI systems

Deep dives

Technical case studies

Domains

CV · Arabic NLP/RAG · MLOps

Applied AI systems

Featured AI Systems

Applied AI projects presented through the problem, implementation, and engineering decisions behind them.

Deployed

Medical Computer Vision · MLOps

Chest X-Ray MLOps System

Analyze chest X-rays for abnormalities through a deployable medical computer vision workflow.

Strongest proof

Grad-CAM explainability output

View Case Study Demo

System boundaries

Model
API
Docker
Deploy
Monitor

Architecture idea

A DenseNet121-powered system with a FastAPI backend, Streamlit frontend, Grad-CAM explainability, and automated deployment.

Technical implementation

Fine-tuned DenseNet121 serves inference and Grad-CAM through FastAPI, while GitHub Actions builds and deploys the Dockerized service and Streamlit interface for cloud delivery.

Stack / tools

PyTorchDenseNet121FastAPIStreamlitDockerAzureGitHub ActionsGrad-CAM

Separates the model service from the UI so the inference backend can evolve independently from the user-facing app.

Supporting links

Code

Deployed

Arabic NLP · Sentiment Systems

JorSenti-LoRA

Handle Jordanian and Levantine sentiment analysis where slang, sarcasm, and neutral sentiment routinely break generic Arabic classifiers.

Strongest proof

LoRA / PEFT adaptation for Jordanian Arabic

View Case Study Demo

System boundaries

Model
API
Docker
Deploy
Monitor

Architecture idea

A sentiment engine for Talabat-style Arabic reviews with LoRA adaptation, weighted loss, and a Hugging Face Spaces demo.

Technical implementation

Adapted MARBERTv2 with PEFT/LoRA, engineered a weighted cross-entropy loss for the neutral class, tracked experiments with MLflow, and packaged the app for Hugging Face Spaces.

Stack / tools

PythonPyTorchMARBERTv2LoRA / PEFTMLflowDockerStreamlitHugging Face Spaces

Uses parameter-efficient fine-tuning so the model can be trained locally while still targeting difficult dialect-specific behavior.

Supporting links

Code

Complete

Arabic NLP / RAG

TaLibAI

Answer Arabic and English study questions from course materials instead of relying on generic model memory.

Strongest proof

Arabic-first grounded Q&A evidence

View Case Study Code

System boundaries

Model
API
Docker
Deploy
Monitor

Architecture idea

A bilingual RAG study assistant with manual FAISS fundamentals, a LangChain refactor, OCR support, and RTL-aware retrieval.

Technical implementation

Built a from-scratch retrieval pipeline with Arabic normalization and cosine search, then re-implemented the system with hybrid FAISS plus BM25 retrieval, LCEL chains, and local Gemma generation.

Stack / tools

PythonFastAPIFAISSLangChainGemmaOllamaGemini Flash OCRStreamlit

Shows both low-level retrieval understanding and framework-level system design in the same repo.

Complete

Computer Vision · Detection & Tracking

Anti-UAV Detection

Detect, classify, and track UAVs in RGB images and video through a full computer vision workflow.

Strongest proof

Detection and tracking demo recording

View Case Study Watch demo

System boundaries

Model
API
Docker
Deploy
Monitor

Architecture idea

A drone detection and tracking system with comparative training, ByteTrack integration, MLflow tracking, and a multi-page Streamlit interface.

Technical implementation

Combined YOLOv11-S, RT-DETR-R50, and ByteTrack, organized repeatable training configs, logged experiments in MLflow, and wired CI/CD plus Docker for reproducible delivery.

Stack / tools

YOLOv11RT-DETRByteTrackMLflowStreamlitDockerGitHub ActionsRuffPytest

Treats model comparison, tracking, and deployment as one system instead of isolated notebooks.

Supporting links

Code

Complete

Information Retrieval · RAG

Mini Search Engine with RAG Extension

Build a classical information retrieval system first, then extend it with grounded answer generation.

Strongest proof

MAP/NDCG evaluation before RAG extension

View Case Study Code

System boundaries

Model
API
Docker
Deploy
Monitor

Architecture idea

A TF-IDF search engine with Cranfield evaluation and a Gemini-powered RAG extension.

Technical implementation

Implemented parsing, preprocessing, inverted indexing, sparse TF-IDF retrieval, MAP/NDCG evaluation, and top-k document handoff into a cited RAG response pipeline.

Stack / tools

PythonTF-IDFCosine SimilarityNLTKSciPyCranfieldGemini 2.5 Flash

Keeps evaluation explicit instead of treating RAG as a black-box improvement.

Mini Search Engine with RAG Extension preview

Complete

Computer Vision · Aerial Surveillance

AerialGuard

Detect and track tiny vehicles and pedestrians in high-altitude UAV imagery where scale variance breaks standard detectors.

Strongest proof

SAHI slicing for tiny aerial targets

Code

Architecture idea

An aerial detection pipeline with SAHI slicing, YOLOv11 inference, tracking, experiment tracking, and a local Streamlit interface.

Technical implementation

Fine-tuned YOLOv11 for aerial scenes, added SAHI for dense small-object inference, integrated Bot-SORT tracking, and organized MLflow, DVC, Docker, and CI workflows around the pipeline.

Stack / tools

YOLOv11SAHIBot-SORTMLflowDVCDockerStreamlitGitHub Actions

Uses slice-aided inference to handle sub-15-pixel targets without treating resolution as the only answer.

Skills system

Tools I use to ship ML systems

A concise view of the stack I rely on across model work, APIs, deployment, and product-facing interfaces.

Computer Vision

Vision models and image-processing pipelines.

PyTorch

training and serving vision models

OpenCV

preprocessing and postprocessing stages

YOLO / Object Detection

detection, tracking, and model comparison workflows

NLP / RAG

Arabic language systems and retrieval-backed flows.

Hugging Face

Arabic model adaptation, demos, and model delivery

FAISS / RAG

retrieval pipelines, Arabic normalization, and grounded responses

AraBERT

dialect-aware sentiment modeling for Arabic review data

MLOps / Backend

Inference APIs, packaging, and deployment paths.

FastAPI

typed model-serving APIs

Docker

reproducible runtime packaging

MLflow

experiment and artifact tracking

Frontend / Product

Clear interfaces for product-facing AI systems.

Next.js / React

portfolio shell and product-facing interfaces

TypeScript / Tailwind

safer UI integration and fast styling systems

Streamlit

fast ML demos and experiment-facing interfaces

Tools / Cloud

Delivery, collaboration, and hosted runtime tooling.

Azure / Deployment

cloud-ready hosting and delivery workflows

MLflow

experiment tracking and artifact comparison

GitHub Actions

CI/CD automation and repeatable delivery

Skills to projects

Where the stack shows up in real work

A recruiter-friendly map from core technical skills to the projects where they are actually used across model work, APIs, deployment, retrieval, and product interfaces.

PyTorch

ModelExplainabilityDeployment

Projects where used

Chest X-Ray MLOps SystemCase Study JorSenti-LoRACase Study Night Vision Image EnhancementSection

FastAPI

Backend

APIInference

Projects where used

Chest X-Ray MLOps SystemCase Study TaLibAICase Study

Docker

MLOps

DockerizedDeploymentCI/CD

Projects where used

Chest X-Ray MLOps SystemCase Study JorSenti-LoRACase Study Anti-UAV DetectionCase Study AerialGuardSection

MLflow

Tooling

Experiment trackingEvaluation

Projects where used

JorSenti-LoRACase Study Anti-UAV DetectionCase Study AerialGuardSection

Hugging Face

NLP/RAG

DeploymentDemo

Projects where used

JorSenti-LoRACase Study

FAISS / RAG

NLP/RAG

RetrievalEvaluation

Projects where used

TaLibAICase Study Mini Search Engine with RAG ExtensionCase Study

OpenCV

Image processing

Projects where used

Night Vision Image EnhancementSection

YOLO / Object Detection

DetectionTracking

Projects where used

Anti-UAV DetectionCase Study AerialGuardSection

Azure / Deployment

MLOps

DeploymentCI/CD

Projects where used

Chest X-Ray MLOps SystemCase Study

Streamlit

Frontend

UIDeployment

Projects where used

Chest X-Ray MLOps SystemCase Study JorSenti-LoRACase Study TaLibAICase Study Anti-UAV DetectionCase Study AerialGuardSection

AI systems workflow

How I build AI systems end-to-end

I treat the model as one part of a complete system, with explicit gates from data ingestion through serving and monitoring.

01
Data Ingestion
Collect project inputs and keep datasets traceable before modeling.
- DVC
02
Validation
Check formats, labels, and pipeline assumptions before training.
- DVC
- GitHub Actions
03
Training
Run model experiments with repeatable configs and tracked results.
- MLflow
04
Evaluation
Compare task quality before moving a model toward serving.
- MLflow
05
Registry / Promotion Gate
Promote only documented artifacts that pass review and delivery checks.
- MLflow
- Docker
06
Serving
Expose inference through a tested API or application interface.
- FastAPI
- Docker
- Streamlit
07
Monitoring
Watch failure modes, latency, and model behavior after release.
- MLflow
- Monitoring

About Me

I am a Computer Science student at Jordan University of Science & Technology (JUST) specializing in building production-ready intelligent systems. I am passionate about the entire machine learning lifecycle—from the mathematical foundation of Deep Learning to architecting MLOps infrastructure.

"Failure never stopped my journey. It sharpened my thinking and discipline."

In Computer Vision, I developed an automated Chest X-Ray clinical-assistance prototype. In NLP, I fine-tuned AraBERT Transformers for Arabic sentiment analysis.

I focus heavily on MLOps—containerizing with Docker, building APIs with FastAPI, and automating deployments on Azure.

View Projects Contact

Career arc

Progression toward AI engineering

A concise view of how my work connects CS fundamentals, applied ML, and production-minded delivery.

01 · Foundation
Artificial Intelligence at JUST
Building the CS fundamentals behind algorithms, software systems, and applied machine learning.
02 · Computer Vision
Medical and aerial vision systems
Applying deep learning to Chest X-Ray analysis and object-detection workflows.
03 · Arabic NLP / RAG
Language systems for Arabic data
Working with AraBERT-style sentiment analysis and retrieval-focused NLP systems.
04 · MLOps
From model to deployable service
Connecting models to FastAPI services, Dockerized runtimes, and deployment-oriented workflows.

Exploration beyond the core stack

Experiments / Creative Engineering

Smaller systems and technical experiments that broaden my product and engineering practice.

Complete

Computer Vision · Image Enhancement

Night Vision Image Enhancement

Technical question: How do low-light enhancement methods compare when improving visibility without ignoring artifact risk?
Objective: Improve visibility, contrast, and illumination consistency in low-light imagery.

Observation

The project frames enhancement as a method-comparison problem rather than assuming one technique wins everywhere.

Low-light preprocessing needs artifact-aware review, not only brighter output.

Code

Method / setup

Compare CLAHE, gamma correction, multi-scale Retinex, color Retinex, and a deep learning enhancer on dark images.

Architecture idea

A comparative enhancement workflow spanning histogram-based methods, Retinex variants, and a learned low-light model.

Technical implementation

Implemented CLAHE, gamma correction, multi-scale Retinex, color Retinex, and a deep learning enhancer to compare their behavior on dark images.

Stack / tools

PythonPyTorchOpenCVRetinexCLAHEMATLAB

Frames enhancement as a method-comparison problem rather than assuming one technique wins everywhere.

Complete

Creative Engineering

3D Donut Engine

Technical question: Can a procedural 3D asset be presented without adding WebGL to the default page path?
Objective: Explore procedural 3D rendering while protecting initial portfolio performance.

Observation

The interactive preview loads only after explicit user intent.

Heavy visual previews should be intent-gated when they are not the primary content.

Code

Method / setup

Build the scene in Blender and place the exported model behind the existing Explore 3D interaction.

Larger-system relevance

It became the portfolio's click-gated 3D preview.

Architecture idea

A Blender scene using geometry nodes, procedural shading, texture work, and particle systems.

Technical implementation

Built the scene in Blender and exposed the exported model as an optional interactive portfolio preview without putting WebGL on the default path.

Stack / tools

Blender 4.2Geometry NodesCyclesThree.js

The WebGL preview loads only after explicit user intent to protect initial performance.

Supporting links

Demo

Complete

Interactive Systems

Wing Jump Arcade Game

Technical question: How can a simple Flappy Bird-style mechanic become a fuller local multiplayer game system?
Objective: Build responsive local arcade gameplay with single-player and two-player modes.

Observation

The project expands a simple arcade premise through multiple control paths and progression logic.

Game feel depends on state boundaries, input handling, and progression as much as sprite movement.

Code

Method / setup

Structure the Pygame loop around states, keyboard and controller input, collision, health, backgrounds, and difficulty progression.

Architecture idea

A Pygame arcade game with single-player and two-player modes, health systems, dynamic backgrounds, and increasing difficulty.

Technical implementation

Structured the game loop around gameplay states, controller and keyboard input, collision handling, health, and difficulty progression.

Stack / tools

PythonPygameOOPGame Design

Expands a simple arcade premise into a fuller local game system with multiple control paths and progression logic.

Supporting links

Demo

In progress

Systems Research

Next-Gen AI Pipeline

Technical question: What boundaries should a reusable AI orchestration pipeline define before implementation?
Objective: Explore architecture for connecting model execution, retrieval, APIs, deployment, and observability.

Observation

The project is intentionally marked in progress rather than presented as completed production proof.

Architecture concepts should be scoped before implementation claims.

Method / setup

Keep the work in research and design while defining system boundaries before implementation.

Architecture idea

An in-progress architecture concept for connecting model execution, retrieval, APIs, deployment, and observability into one reusable pipeline.

Technical implementation

Currently in research and design. The goal is to define clean system boundaries before implementation rather than presenting an unfinished system as production-ready.

Stack / tools

ResearchArchitectureModel OrchestrationComing Soon

Kept intentionally marked as in progress so the portfolio separates completed proof from active research.

Certifications & Credentials

5 verified credentials ●

Issuing Authority

DeepLearning.AI & Stanford

4 credentials issued by DeepLearning.AI and Stanford University via Coursera. Click any badge to view the official verified record.

4 verified with direct credential links

Verified credential

DeepLearning.AI & Stanford

Verified

Machine Learning Specialization

Issued: Issued Jul 2, 2025
Issuer: DeepLearning.AI & Stanford

Verify credential

Verified credential

DeepLearning.AI & Stanford

Verified

Unsupervised Learning, Recommenders, Reinforcement Learning

Issued: Issued Jul 2, 2025
Issuer: DeepLearning.AI & Stanford

Verify credential

Verified credential

DeepLearning.AI & Stanford

Verified

Advanced Learning Algorithms

Issued: Issued Jun 29, 2025
Issuer: DeepLearning.AI & Stanford

Verify credential

Verified credential

DeepLearning.AI & Stanford

Verified

Supervised Machine Learning: Regression and Classification

Issued: Issued Jun 15, 2025
Issuer: DeepLearning.AI & Stanford

Verify credential

4 CREDENTIALS SHOWN●5 DIRECT LINKS

Get In Touch

Available for internships, junior AI/ML roles, computer vision work, RAG or NLP projects, and backend ML API implementation. If you need someone who can move from model prototype to deployable system, contact me here.

Open toInternshipsComputer VisionRAG / NLPBackend ML APIs

Best fit

Internships

AI, ML, or software roles with real shipping exposure.

Computer Vision

Inference pipelines, image enhancement, and model-serving work.

RAG / NLP

Arabic NLP, retrieval workflows, and applied language systems.

Backend ML APIs

FastAPI services, Dockerized inference, and deployment paths.

Junior AI/ML Roles

Teams that need hands-on implementation across model and product layers.

Email

aljafariabdullah2005@gmail.com

The fastest route for hiring conversations, project inquiries, or collaboration on applied AI systems.

AI Engineer buildingdeployable ML systems.

Featured AI Systems

Chest X-Ray MLOps System

JorSenti-LoRA

TaLibAI

Anti-UAV Detection

Mini Search Engine with RAG Extension

AerialGuard

Tools I use to ship ML systems

Where the stack shows up in real work

PyTorch

FastAPI

Docker

MLflow

Hugging Face

FAISS / RAG

OpenCV

YOLO / Object Detection

Azure / Deployment

Streamlit

How I build AI systems end-to-end

Data Ingestion

Validation

Training

Evaluation

Registry / Promotion Gate

Serving

Monitoring

About Me

Progression toward AI engineering

Artificial Intelligence at JUST

Medical and aerial vision systems

Language systems for Arabic data

From model to deployable service

Experiments / Creative Engineering

Night Vision Image Enhancement

3D Donut Engine

Wing Jump Arcade Game

Next-Gen AI Pipeline

DeepLearning.AI & Stanford

Machine Learning Specialization

Unsupervised Learning, Recommenders, Reinforcement Learning

Advanced Learning Algorithms

Supervised Machine Learning: Regression and Classification

Get In Touch

Best fit

Email