AbdulRahman Adham Rashed
AI / Machine Learning EngineerData science, NLP & generative AI
I build intelligent systems that turn complex information into decisions you can trace, review, and explain.
Keep the evidence attached.
Retrieve relevant passages within the user’s access scope. Carry source references forward so the answer can be checked against the original.
Interactive concept illustration
Selected engineering
Explore the systems, the decisions,
and the evidence behind the work.
Document intelligence
Enterprise Document Intelligence
From English and Arabic documents to reviewed business data, cited policy evidence and explainable decisions.
Local development platform
Applied multi-agent systems
Multi-Agent Training Operations
Five connected academy workflows, grounded in approved course knowledge and coordinated around explicit human review.
Team project · Andalusia Academy
An experiment in efficiency
Implementing low-rank adaptation directly in PyTorch, then comparing parameter efficiency on EuroSAT’s ten land-use classes.
Explore the LoRA case study| Method | Trainable | Accuracy |
|---|---|---|
| Full fine-tuning | 100.00% | 98.74% |
| LoRA r=64 | 0.69% | 98.13% |
Single run per configuration · 5 epochs · 80/20 split
How I think about engineering
I’m an Information Systems graduate who learns by building and testing. My work connects machine learning, retrieval and agent workflows with the backend engineering that makes them useful.
From supervised learning and transformer fine-tuning to complete AI applications, I care about what happens before and after a model responds.
Models propose. Evidence, deterministic logic and human review determine authoritative outcomes.
Project archive
From a first baseline to the systems around it.
Eight projects across AI, data and software.
8 projects in the archive
Machine Learning
Academic project · supplied results
7,043 telecom records. Random Forest accuracy 76.7%, ROC-AUC 0.842, cross-validation F1 0.634. Clustering on 10,000 Census records, DBSCAN silhouette 0.297.
Reported in supplied background. Source artifacts are not linked.
NLP
Academic project · supplied results
Restaurant-domain intent accuracy 93.3%, approximately 0.93 macro-F1. Paraphrasing, slot extraction and structured JSON.
Reported in supplied background. Source artifacts are not linked.
AI Systems
Repository project
Structure-aware PDF/DOCX/TXT/Markdown ingestion, persistent dense retrieval and source-grounded generation.
No BM25 hybrid search, reranker or evaluation set.
AI Systems
Repository project
Parser, scorer and critic roles, duplicate detection, evidence-based ranking and human approval before invitation actions.
Single-model correlated errors, exact-match deduplication, no completed fairness audit or automated tests.
Data Science
Academic project · supplied results
10,127 records. Naive Bayes accuracy 87.3%, balanced accuracy 74.9%. Classification, DBSCAN and dashboard.
Reported in supplied background. Source artifacts are not linked.
Computer Vision
Academic project · supplied results
Scratch CNN vs transfer learning, augmentation and background-removal experiments. Transfer-learning accuracy 92.31% on only 13 held-out images.
Reported in supplied background. Source artifacts are not linked.
Analytics
Academic project · supplied results
Weather exploration and dashboards using 145,460 records.
Reported in supplied background. Source artifacts are not linked.
Backend / Data Engineering
Academic project · supplied results
Normalization, foreign keys and query restructuring. One reported retrieval example reduced from six joins to two.
Reported in supplied background. Source artifacts are not linked.
Research & experiments
Smaller studies in adaptation, quantization and deployment. Academic designs and exploratory work are labelled by scope.
Descriptions and parameter counts come from supplied academic background. Further numeric results await source artifacts.
Prompt Tuning, Prefix Tuning, P-Tuning, BitFit, IA3 and DistilBERT. Reported baseline ~66.96M trainable parameters and BitFit ~52,226 (0.078%).
PTQ, QAT, INT8 calibration and KL-divergence analysis. Model size, accuracy and latency comparisons need source artifacts before publishing numeric claims.
BentoML single/batch inference, Docker and responsiveness experiments.
Training and internship management concept with verified opportunities, applications, training-hour tracking and administrator approvals.
ERP/BPMN logistics workflows, responsibilities, decision gateways and DFD analysis.
Engineering capabilities
Practical depth grows through projects.
Each group states where the experience comes from.
Applied in featured projects
Grounded generation, agent orchestration, source verification, human review and evaluation.
Project and experimental practice
Classification, regression, clustering, feature engineering, cross-validation and transformer adaptation.
Applied in projects
APIs, relational storage, tests, containers and observable workflows that make model behavior inspectable.
Academic and project practice
Data preparation, exploratory analysis, segmentation and interactive reporting.
Additional coursework and exercises Java · C++ · MongoDB · BentoML
Education
September 2026
Arab Academy for Science, Technology & Maritime Transport
University of Northampton dual degree partner programme. GPA 3.1/4.0, Very Good with Honors.
From supervised learning and analytics to transformer fine-tuning, retrieval and complete AI applications.
Let’s make something meaningful
AI/ML engineering opportunities, research
collaborations, or a thoughtful technical question.