Senior AI Engineer with 5+ years across LLM training pipelines, ML research, automation engineering, and mentorship. I turn complex technical problems into reliable systems with clear business value, and I help other engineers do the same.
Not just impressive demos, measurable, maintainable, useful in the real world.
From data and model evaluation to deployment, automation, and the people who maintain it. Five ways I help teams ship.
Fine-tuning workflows (LoRA/QLoRA, RLHF), retrieval pipelines, and production serving with vLLM, Ollama, and llama.cpp, plus self-correcting agent stacks like Hermes. Built for reliability, measurement, and useful output.
Taking ML systems from prototype to production with strong engineering practices, cloud infrastructure, and clear quality checks.
Workflow automation that removes repetitive manual work across data, operations, research, and communications, at enterprise scale.
Turning research ideas into working systems, then explaining the tradeoffs clearly to technical and non-technical stakeholders alike.
Helping teams reason through architecture, evaluation, deployment, and product fit, and mentoring engineers toward their first production deployment.
Production ML, published research, AI automation, and enterprise systems. Some clients are confidential, framed by capability, never identity.
A production-grade retrieval system for legal documents, built to minimise hallucinations and maximise citation-backed accuracy. I lifted it from 29% to 96.7% by rebuilding the evaluation methodology, not by chasing surface-level model swaps. Designed for domains where trust, traceability, and correctness are non-negotiable.
Published in Franklin Open (Elsevier). A Cross-Domain Ensemble Probability Fusion framework achieving 94.4% cross-dataset accuracy on heterogeneous data with zero feature overlap, a 64.3% relative gain over baseline.
Privacy-preserving federated learning across 15 distributed contractors, 91.4% R² accuracy with no raw data ever leaving its source. Demonstrates ML engineering under strict privacy and governance constraints, with significant cost-savings potential.
ML-powered communication intelligence that digests emails and meeting transcripts, surfaces key themes via HDBSCAN clustering and embeddings, and turns them into structured weekly briefs, orchestrated with n8n.
AI-driven automations that remove manual work across email campaigns, research, and operations, built on n8n (drawing from a library of 4,000+ workflow patterns), Google Apps Script, and agentic LLM pipelines that self-route, enrich, and summarise data end to end.
NLP research benchmarking ClinicalBERT against TF-IDF across 49,810 patient records for diagnosis classification, examining where transformer models earn their cost in clinical text and where classical methods still hold.
Quantum kernel methods for cross-dataset phishing detection, simulated on NVIDIA H100. 90.3% mean cross-dataset accuracy (AUC 0.967) where all classical transfer-learning baselines collapse to near-random. Under review at an Elsevier venue.
An end-to-end travel and operations management system built on Google Apps Script, automating multi-stage approval workflows for an enterprise team. Shows automation engineering that removes manual overhead at organisational scale.
A full-stack SaaS learning-management platform with billing, onboarding, and payment integration, currently in stealth with customers waiting. Architected for multi-tenant scale; details under wraps until launch.
I do my best work where experimentation meets execution.
I am a Senior AI Engineer, machine learning researcher, and recognised mentor focused on building AI systems that work in the real world, and developing the people who build them. I enjoy the part of engineering that turns a promising prototype into something dependable: evaluation, deployment, monitoring, iteration, and maintenance.
At Turing, I work on LLM training pipelines, RLHF, and supervised fine-tuning, sharpening how I reason about model quality, data curation, and alignment. At Start Innovation Hub Nigeria, I build end-to-end ML pipelines, automate high-volume workflows, and mentor data professionals; that work earned a Staff of the Year award twice. At Visis Startup, I built backend ML infrastructure for an accessibility platform serving thousands of visually impaired users.
My research matters to me too. I've published in cross-dataset phishing detection through Franklin Open (Elsevier, 2026) and contribute to work on quantum kernels and explainable healthcare analytics. Research gives me a disciplined way to think about uncertainty, generalization, and honest measurement, the same standard I bring to production systems.
One published in an Elsevier venue, several in active peer review. My research spans cybersecurity, public health ML, and biostatistical epidemiology.
The Cross-Domain Ensemble Probability Fusion framework, PCA-based feature harmonisation and information-theoretic weighted fusion for 94.4% cross-dataset accuracy across heterogeneous datasets with zero feature overlap.
DOI: 10.1016/j.fraope.2026.100579 · Open Access ↗Systematic investigation of quantum kernel methods for cross-dataset phishing detection, simulated across 8 qubit configurations on NVIDIA H100.
Deeper analysis of qubit scaling, kernel sparsity phenomena, and the practical design guideline 2q ≪ n for quantum kernel transfer learning.
Multi-country ML study across Benin, Gambia, Ghana, and Nigeria, 16,852 adolescent girls. LightGBM, SMOTE, Boruta, and LIME on the most recent DHS data.
Survey-weighted LightGBM on 5,998 facility-delivered women from the 2024 NDHS, with SHAP explainability and calibration analysis.
ClinicalBERT vs TF-IDF benchmarking across 49,810 patient records, examining where transformer models earn their cost in clinical text.
Bivariate copula regression with district-specific parameters and Markov random fields; GLMM modelling across Ghana, Gambia, Rwanda, and Uganda DHS/MIS data.
DHS data analysed across Nigeria · Ghana · Gambia · Rwanda · Uganda · Benin
Awarded twice for measurable engineering and organisational impact.
Generative AI Track. Recognised by Dr. Olubayo Adekanmbi and Dr. Roti Balogun.
Cohort 2, 2025, recognised for impactful mentorship delivery.
Selected from 12,101+ completed applications.
Top 30 Finalist out of 9,000+ applicants (0.3% selection rate), 2024.
Peer-reviewed publication in Franklin Open, with further papers in review.
Saviour guided fellows through their AI learning journeys and delivered impactful sessions in the Generative AI Track, strengthening their confidence for the next phase of their careers.
Recognised as a Spotlight Mentor for active engagement and the delivery of impactful sessions across the DeepTech Ready programme.
Selected from over 12,000 applicants into the Builder Track, recognition of the kind of production-minded engineering this portfolio is built on.
Whether you need a senior AI engineer, a research collaborator, or someone who can take your ML prototype to production, I'd like to hear about it.