Omar Bin Talib | AI/ML Engineer & Data Scientist

01 // About

BACKGROUND

I'm Omar Ali Bin Talib, a computer science student at King AbdulAziz University with a 4.9/5.0 GPA, working in machine learning, computer vision, and data engineering.

I completed training at the KAUST AI Academy (top 1% of applicants) and conducted research at the University of Toronto's Data Sciences Institute. My recent work includes Bidaya, a source-grounded AI map of the Prophet's life, psychometric simulations with large language models, object detection models for Kaggle competitions, automated validation tools for census data, and an Arabic-first mental health intake that placed in the top 5 at TCS Sustainathon Saudi Arabia 2026.

02 // Projects

CODE AND COMPETITIONS

Featured Project

بداية (BIDAYA): An AI-Powered Map of the Sirah

The life of the Prophet ﷺ told as one story on an interactive map: 142 events in place and in order, the verses revealed about them, and 98 people, in Arabic and English. Its assistant, “Ask the map”, answers only from approved sources, cites them, and moves the map to the event the answer rests on.

142 Events On the Map

2,905 Approved Source Passages

97.9% Answers With a Correct Source

Expand Preview

Built a RAG pipeline in which an LLM router classifies each question (Sirah, verse, companion, fatwa, personal, off-topic), refuses fatwa and personal ones, and rewrites follow-ups into standalone Arabic and English search queries.
Hybrid retrieval (bge-m3 dense + BM25, fused with RRF) over 2,905 passages from Dorar, the Sahihayn and Ar-Raheeq Al-Makhtum, re-ranked by bge-reranker-v2-m3; answers must cite the passages or are replaced by a fixed "no source" reply.
Evaluated on 60 benchmark and 50 adversarial questions: 97.9% sourced answers, 100% correct refusals, no false statements on the hard set (a general chatbot without sources erred in 8 of the same 50).
Served by a FastAPI backend on Modal (T4 GPU, scales to zero), with a React + TypeScript map frontend on Vercel and pre-recorded Arabic/English narration for all 142 events.

RAG bge-m3 + BM25 Reranking FastAPI Modal GPU React / TypeScript Arabic NLP

Project Details Live Demo ↗

3rd / 130+ Teams Team Optimizers

حريص (HARIS): Semantic Data Validation Engine

GASTAT Data Innovation Hackathon (General Authority for Statistics & SDAIA)

Automated validation service for national census datasets that checks cross-field logical consistency across 400-column tables. Built with Hassan Algamdi and Yasser Albogami, finishing 3rd out of 130 teams.

12× Latency Reduction (60s → 5s)

160+ Business Rules Automated

~30s Batch Audit (400+ Cols)

Expand Preview

Built an asynchronous FastAPI service to run validation checks concurrently, reducing batch latency from 60 seconds to 5 seconds.
Used structured JSON outputs with Pydantic and GPT-4o to enforce statistical validation rules.
Wrote custom validation logic for household relationships, employment status, and demographic consistency.

FastAPI GPT-4o Pydantic Asyncio Data Validation

Project Details GitHub Repository ↗

3rd / 100+ Teams (Kaggle) Team Sekiro

DAL Shemagh Detection Challenge

Two-Stage Object Detection Pipeline

Computer vision pipeline built for the DAL Shemagh Challenge on Kaggle. Placed 3rd on the private leaderboard with a 0.863 score, within 0.03 of 1st place among more than 100 teams.

0.863 Private Leaderboard Score

5 Samples Initial Positive Labels

100+ Competing Teams

Expand Preview

Combined YOLO11m for initial bounding box proposals with a ConvNeXt-Tiny classifier to filter false positive detections.
Expanded five initial labeled images through an iterative pseudo-labeling loop on unlabeled test imagery to increase training data.
Applied HSV photometric augmentation and Weighted Boxes Fusion (WBF) to merge overlapping candidate boxes.

PyTorch YOLO11m ConvNeXt WBF Ensembling Computer Vision

Project Details Kaggle Solution Writeup ↗

Applied AI Platform Data Managing Center · KAU

Survey Pilot

Synthetic Survey Piloting and Scale Validation

Platform for testing research questionnaires with persona-conditioned language model respondents. Calculates Cronbach's alpha reliability, bootstrap confidence intervals, and item correlations before deploying surveys to human test groups.

≤ 0.05 Baseline Error (Cronbach's α)

5 Styles Simulated Archetypes

9-Step Scale Extraction Pipeline

Expand Preview

Conditioned synthetic respondents on demographic backgrounds and five response styles (balanced, acquiescent, extreme, midpoint, and careless) to counter agreement bias.
Extracted research scales from OpenAlex and stored high-dimensional vector embeddings in PostgreSQL with pgvector.
Built a FastAPI backend coordinating local Ollama instances for survey runs and vLLM on Modal for large batch extraction jobs.

FastAPI vLLM / Modal Ollama pgvector PostgreSQL Psychometrics

Project Details

Top 5 / ~500 Registrations Team Mentalicious

هدوء (HUDOU): AI Mental Health Intake

TCS Sustainathon Saudi Arabia 2026 (AI for Community Health & Well-being)

Arabic-first mental health intake that starts at the physical symptom people are willing to name and hands the clinician a one-page structured summary. Built with Elyas AlTurkistani and Sulaiman AlAngari; placed in the top 5 of ~500 registrations at the grand finale in Riyadh.

Top 5 From ~500 Registrations

2 Screens Patient & Clinician Views

4-Step Privacy Ladder

Expand Preview

Starts from a physical complaint and rules out medical causes (fever, weight loss, sudden pain or numbness) before any psychological path, routing those cases to a regular doctor.
Runs PHQ-9 and GAD-7 screening as everyday conversation in Saudi dialect, while the clinician console turns the same words into a structured record with an urgency grade and a queue sorted by risk.
Designed a safety layer with a separate watchdog model and a fixed risk-phrase list that halts the chat and escalates to a human, plus a four-step privacy ladder that reveals identity only when the user chooses.

Conversational AI Arabic / RTL PHQ-9 · GAD-7 AI Safety Digital Health

Project Details

Open Source Research University of Toronto · DSI

Studying Numerical Literacy Across Generations in U.S. Households

Longitudinal Econometric and Machine Learning Analysis (1970 - 2025)

Empirical study evaluating how mathematical literacy passes between generations and affects long-term wealth, using 55 years of survey data from the Panel Study of Income Dynamics (PSID).

~0.80 Model R² Score

50,000+ Household Records

55 Years Longitudinal Waves (PSID/CDS)

Expand Preview

Ingested, cleaned, and merged 55 years of household survey records from the Panel Study of Income Dynamics (PSID) and Child Development Supplement (CDS).
Trained Random Forest and XGBoost regression models to measure how childhood math scores predict adult net worth, reaching an R² of 0.80.
Adjusted for longitudinal survey attrition and sampling weight shifts across multiple waves.

Python Pandas Random Forest XGBoost Econometrics PSID / CDS

Project Details GitHub Repository ↗ Showcase Poster (PDF) ↗

03 // Experience & Education

WORK, RESEARCH, AND ACADEMICS

07/2026 - 09/2026

King AbdulAziz University

Data Managing Center

AI Engineer · Internship

Built a synthetic-respondent pilot engine using persona-conditioned language models to calculate psychometric reliability statistics, including Cronbach's alpha, bootstrap confidence intervals, and item-total correlations.
Benchmarked synthetic pilot responses against human baseline data, estimating reliability within 0.05 Cronbach's alpha error of real survey results.
Created an automated review layer to flag double-barreled questions, leading wording, and ambiguous items before running full pilots.
Engineered a FastAPI backend orchestrating local Ollama runs and vLLM batch jobs on Modal, with a 9-step pipeline that extracts validated scales from OpenAlex into PostgreSQL with pgvector.

06/2025 - 08/2025

University of Toronto

Data Sciences Institute (DSI)

AI Researcher · SUDS Fellow

Completed a 440-hour research placement at the Data Sciences Institute as part of the KAUST Academy AI Specialization Track.
Extracted, cleaned, and harmonized 55 years of longitudinal survey data from the Panel Study of Income Dynamics (PSID) and Child Development Supplement (CDS), covering 50,000+ household records and resolving cross-wave schema changes, attrition, and sampling-weight shifts.
Trained Random Forest and XGBoost regression models to evaluate the relationship between childhood math skills and adult wealth accumulation, reaching an R² of approximately 0.80.
Presented findings at the SUDS 2025 Research Showcase to 30+ faculty and researchers.

10/2024 - 08/2025

Education

King Abdullah University of Science and Technology

KAUST Academy · NTDP

AI Specialization Track

Top 1% Nationwide: 1 of 10 accepted from 17,000+ applicants

Completed a four-stage curriculum covering mathematical foundations of machine learning, deep learning architectures, computer vision, and AI research methods.
Earned specialization credentials across foundational machine learning theory, deep learning implementations, and applied research applications.

2023 - 06/2027 (Expected)

Education

King AbdulAziz University

Faculty of Computing and Information Technology

B.S. in Computer Science

Cumulative GPA: 4.9 / 5.0

Rigorous core curriculum in data structures, design and analysis of algorithms, discrete mathematics, operating systems, and database systems.
Consistently maintained highest honors standing across both foundational computing and applied artificial intelligence coursework.

04 // Skills

LANGUAGES, FRAMEWORKS, AND TOOLS

Core Stack

Programming Languages

Python SQL Java

Intelligence & Vision

Machine Learning & AI

PyTorch Scikit-learn Computer Vision Large Language Models (LLMs) NLP Agentic AI Prompt Engineering

Data & Insights

Data Science & Analytics

Pandas NumPy Feature Engineering EDA Matplotlib Seaborn Data Visualization Data Pipelines & Cleaning

Development & APIs

Tools & Infrastructure

FastAPI Pydantic LangChain Ollama PostgreSQL Jupyter Git & GitHub

05 // Licenses & Certifications

CREDENTIALS

KAUST Academy · NTDP

3 Verified Credentials

Artificial Intelligence Specialization & Research Track

10/2024 - 08/2025 · Thuwal, Saudi Arabia and Toronto, Canada

AI Program at the University of Toronto

440 hours of research internship experience in AI laboratories (June 3 - August 19, 2025).

Advanced Artificial Intelligence Training Program

40 hours of coursework in convolutional architectures, deep learning models, and computer vision (February 2025).

Introduction to Artificial Intelligence Training Program

Foundational machine learning, probabilistic modeling, and neural networks (November 2024).

DeepLearning.AI

Specialization

Mathematics for Machine Learning and Data Science

Issued Dec 2024 · Andrew Ng / DeepLearning.AI

Linear Algebra for Machine Learning

Vector spaces, linear transformations, eigenvalues, eigenvectors, and singular value decomposition (SVD).

Calculus for Machine Learning & Data Science

Multivariate calculus, gradients, Jacobians, Hessians, and loss optimization algorithms.

Probability & Statistics for Machine Learning

Probability distributions, Bayesian inference, maximum likelihood estimation (MLE), and hypothesis testing.

Tata Consultancy Services

Certificate of Achievement

Top Finalist · TCS Sustainathon Saudi Arabia 2026

2026 · Hudou, Team Mentalicious · King AbdulAziz University

View Certificate

Get in touch

Open to machine learning engineering roles, research collaborations, and technical projects. Send a message below and I will get back to you soon.

Email Address

omar.a.bintalib@gmail.com

Phone & WhatsApp

+966 56 512 3153

Location

Jeddah, Saudi Arabia · Open to Relocate

Curriculum Vitae

Download CV (PDF)

Professional Profiles