I am driven by curiosity and the challenge of turning bold ideas into working AI systems
Master of Science in Data Science and Analytics – Sept 2024 to Apr 2026 (expected)
Research Focus: AI Safety and Hallucination Mitigation in Healthcare Applications
Thesis: Retrieval Augmented Reasoning with Self-Correction for Biomedical Question Answering. Focus on LLM fine-tuning, retrieval systems with FAISS and Redis, synthetic data generation, and deployment on Hugging Face Inference Endpoints.
Bachelor of Engineering in Computer Engineering – February 2022
Research Focus: Network Systems and Routing Protocols in Intermittent Connectivity Environments
Thesis: Enhanced PRoPHET Routing Protocol with Buffer Management Techniques in DTN. Focus on implementing and analyzing four buffer management techniques (MOFO, DLA, DL, FIFO) with PRoPHET routing protocol in Delay Tolerant Networks using ONE Simulator.
Jun 2024 to Nov 2024
Co-developed ChatJourno, a retrieval-augmented fact-checking assistant for West African journalists. Built scraping pipelines using BeautifulSoup and curated datasets from 50+ trusted news sources for reliable retrieval-augmented generation. Successfully deployed the system to production at chatjourno.com, enabling real-time fact-checking support for newsroom operations.
Jan 2024 to Jul 2024
Standardized healthcare data for EMR migration across two hospitals. Led implementation using Agile methodologies and trained over 100 staff, reducing support issues by 70% through comprehensive change management and technical documentation.
Type: Master's Thesis
Goal: Eliminate AI hallucinations in biomedical question answering by architecting retrieval-augmented reasoning with self-correction mechanisms
Role & Team: Project Lead - orchestrated comprehensive literature review, engineered LLM finetuning pipeline with quantization, architected Redis/FAISS integration, deployed production-ready HIPAA-compliant system
Tools & Methods: Llama-3.1-8B, SciBERT, Flan-T5, FAISS, Redis, Hugging Face Inference Endpoints, PyTorch, synthetic data generation, LLM-as-a-Judge evaluation
Outcome & Impact: Deployed production-ready system achieving 87% error detection accuracy and 0.82 semantic preservation F1-score, establishing new benchmark for healthcare AI safety
Click the poster to zoom in, or use the controls to navigate
Type: Fellowship Project (Centre for Journalism Innovation and Development)
Problem & Goal: In West African newsrooms, journalists face a 15-minute window to update breaking news with credible backstories. Under this pressure, research often comes from unreliable or incomplete sources, leading to inaccuracies. The project aimed to empower journalists with an AI-assisted fact-checking and retrieval system that delivers trustworthy backstories in real time.
Role & Team: AI Fellow – collaborated with journalists and technologists to design and prototype ChatJourno. Built data collection pipelines with BeautifulSoup, engineered the RAG system architecture, curated datasets from trusted sources and newsroom archives, and prototyped the chatbot interface for seamless newsroom integration.
Tools & Methods: BeautifulSoup, Python, Streamlit, RAG pipeline design, dataset curation from trusted sources and news archives
Outcome & Impact: Delivered a functional chatbot that reduces newsroom research time, improves reporting accuracy, and strengthens credibility by providing journalists with reliable, ready-to-use backstories for breaking news.
Type: Personal Project
Goal: Fine-tune Llama-3.1-8B to generate creative, brandable domain names with built-in safety filtering and comprehensive evaluation framework
Role & Team: Individual contributor - conducted dataset creation using hybrid Claude API and manual approach, implemented LoRA fine-tuning with 4-bit quantization, developed systematic evaluation methodology with LLM-as-a-Judge, deployed production model on Hugging Face
Tools & Methods: Llama-3.1-8B-Instruct, LoRA fine-tuning, PyTorch, Hugging Face Transformers, Weights & Biases for experiment tracking, Claude API for dataset generation, systematic quality assessment framework, production deployment
Outcome & Impact: Achieved 99.1% success rate, 100% safety compliance on test set, 82% safety performance on extended testing, deployed live demo with creative domain generation capabilities (e.g., "fitwise.app" vs generic "fitness.com")
Type: Academic Team Project
Goal: Build a production-ready ML system for JB Link telecom to predict customer churn and reduce the 43% quarterly churn rate through proactive retention strategies
Role & Team: Team Lead (Team AbHe-ViPa, 5 members) - architected microservices system, implemented FastAPI model serving, designed automated data pipelines with Airflow, configured monitoring with Grafana and Teams alerts
Tools & Methods: Docker, FastAPI, Streamlit, PostgreSQL, Apache Airflow, Grafana, scikit-learn, microservices architecture, automated data quality validation
Outcome & Impact: Deployed enterprise-grade system with batch prediction optimization, automated workflows (1-2 minute intervals), real-time monitoring dashboards, and comprehensive data quality validation with 7 error detection types
Type: Academic/Personal Project
Goal: Build a fully event-driven, scalable pipeline for processing and analyzing large volumes of transportation data
Role & Team: Individual contributor - architected event-driven ETL pipeline, implemented PySpark transformations, configured fine-grained IAM permissions and encryption, automated job orchestration with CloudFormation
Tools & Methods: AWS Lambda, AWS EMR, AWS Glue, AWS Athena, AWS CloudFormation, PySpark
Outcome & Impact: Processed 1.4M+ records with secure IAM policies, encryption, and automated orchestration
Click to view full-size architecture diagram
Type: Academic/Personal Project
Goal: Build a fault-tolerant, scalable platform for analyzing student performance data with interactive visualizations and high availability
Role & Team: Individual contributor - designed AWS architecture with custom VPC, implemented Streamlit dashboard, configured Auto Scaling Groups and Load Balancer, deployed with high availability across multiple EC2 instances
Tools & Methods: AWS (VPC, EC2, ALB, Auto Scaling, S3, IAM), Streamlit, Python data science stack, interactive data visualization, fault tolerance design
Outcome & Impact: Deployed production-ready system with 2-4 auto-scaling EC2 instances, 70% CPU threshold scaling, comprehensive analytics dashboard with dynamic filtering and correlation insights
Click to view full-size architecture diagram
Interactive dashboard with real-time filtering, performance analysis, and data visualization
Type: Academic Project (Computer Vision Course, EPITA)
Goal: Conduct systematic analysis of Stable Diffusion parameters to understand trade-offs between generation quality, speed, and prompt adherence in text-to-image synthesis
Role & Team: Individual researcher - designed controlled experiments across multiple parameter dimensions, conducted comparative analysis of schedulers, CFG scales, and inference steps, developed optimization methodology for architectural image generation
Tools & Methods: Stable Diffusion, Google Colab, multiple schedulers (DPM, EulerA, DDIM, LMS), systematic parameter optimization, prompt engineering techniques, quality assessment frameworks
Outcome & Impact: Identified optimal parameter combinations (CFG 15.0, 30-40 steps, EulerA scheduler), discovered quality plateau at 40 steps with diminishing returns beyond, established best practices for architectural image generation with photorealistic results
Type: Academic Project (Computer Vision Course, EPITA)
Goal: Implement a Generative Adversarial Network to generate realistic handwritten digits from random noise vectors using adversarial training
Role & Team: Individual contributor - implemented generator/discriminator architectures, conducted adversarial training over 30 epochs, analyzed training dynamics and loss patterns for optimal balance
Tools & Methods: TensorFlow/Keras, GANs, Conv2D/Conv2DTranspose layers, BatchNormalization, LeakyReLU activation, adversarial training dynamics
Outcome & Impact: Successfully generated high-quality MNIST digits, achieved balanced discriminator/generator loss (0.60-0.63 vs 0.89), demonstrated clear progression from random noise to realistic handwritten digits
Type: Academic Project
Goal: Develop object detection and segmentation systems for trash classification and plant disease diagnosis
Role & Team: Individual contributor - built end-to-end pipelines for both detection and segmentation tasks, conducted systematic model tuning, implemented comprehensive evaluation frameworks
Tools & Methods: Ultralytics YOLOv8, PyTorch, evaluation frameworks
Outcome & Impact: Achieved 85.9% mAP on trash detection and 36.6% mask mAP on plant disease segmentation through systematic tuning
Python: PyTorch, TensorFlow, Keras, Scikit-Learn, Hugging Face Transformers, OpenCV, Matplotlib
Other: R, SQL
LLM finetuning (LoRA/QLoRA), CNNs, GANs, object detection, segmentation, Stable Diffusion, data generation and data pipelines
LLM-as-a-Judge, model benchmarking, prompt engineering, metric analysis (ROUGE, BLEU, BERTScore)
Docker, Airflow, Git, Hugging Face, AWS, Streamlit, Redis, CI/CD fundamentals (GitHub Actions), Streamlit dashboards, Tableau, Dataiku, Grafana
PostgreSQL, MySQL, Redis, MongoDB
2023
Comprehensive certification covering data collection, transformation, and organization; data analysis and visualization; and data-driven decision making.
2023
Professional certification in project management fundamentals, including project planning, execution, and stakeholder management.
In Progress - 2025
Currently preparing for AWS Machine Learning Engineer Associate certification covering data preparation for ML, model training and deployment, MLOps workflows, and machine learning infrastructure management on AWS.
Jun 2024 to Nov 2024
Selected from over 500 applications across Nigeria and Ghana for the first AI-in-Journalism Fellowship in Africa. Co-developed ChatJourno, a retrieval-augmented fact-checking assistant for West African journalists, successfully deployed to production.
Sep 2024
Associated with: EPITA: École d'Ingénieurs en Informatique
Awarded a highly competitive postgraduate scholarship by the Nigerian government to pursue advanced studies abroad. Selection based on academic excellence, industry-relevant experience, and commitment to contributing to Nigeria's energy sector.
Duke University, Social Science Research Council, Mathematica · 2023
Selected for participation in SICSS-Calabar, one of only two SICSS locations in Africa. The institute covered advanced modules in data modeling with R, text analysis and web scraping, social science predictive analytics, and research mentorship. Program fully sponsored by Prof. Christopher Bail, Social Science Research Council, Mathematica, Duke University, and Institute for the Study of Societal Issues.
Mar 2018
Associated with: Ahmadu Bello University
Awarded by NNPC, TotalEnergies, and partners to recognize academic excellence and support undergraduate education in Nigeria. Granted to high-performing students in tertiary institutions through a highly selective annual process aimed at fostering talent and manpower development.