Saheed Faremi

Experience

1. PhD researcher · Artificial Intelligence and Cognitive Load Research Lab (AICL), University College Cork

   Present

   Doctoral research on segmenting EEG into microstate sequences with variational autoencoders, supervised by Luca Longo. Builds generative models, large-scale evaluation sweeps, and temporal post-hoc analyses over resting-state EEG.

   • Developed Conv-VaDE, a convolutional variational autoencoder with a learnable Gaussian-mixture latent prior and four polarity-invariance mechanisms, for segmenting EEG topographic maps into microstate sequences
   • Ran a 4,832-model architecture search (486 configurations per participant across state count, latent dimension, depth, and width), scoring 110 metrics per model with cross-subject ICC consistency analysis on SLURM GPU and IBM Power9 clusters
   • Developed Graph-VaDE, a graph-neural-network VAE with a learnable Gaussian-mixture prior that discovers microstates directly on the electrode graph via static anatomical adjacency, replacing interpolated topographic images; recovers ground-truth maps on a synthetic gate benchmark (ARI 1.0, GEV 0.85)
   • Built a post-hoc hidden semi-Markov backfitting stage that learns per-state dwell-time distributions and transition probabilities, then relabels full recordings coherently
   • Quantified recurrence in resting EEG (single-participant case study): the multichannel signal recurs well above phase-randomised surrogates (determinism 0.83 against 0.60) while the discrete label sequence does not exceed a first-order Markov surrogate

2. Data scientist → Founding engineer · Etihuku (Curnance)

   2021-05 → present

   Data science and engineering at Etihuku, a data analytics consultancy, including founding-engineer ownership of Curnance, its multi-asset fintech venture.

   • Founding engineer for Curnance, the multi-asset fintech venture built within Etihuku: shipped seven production services covering wallet/ledger, auth, tiered KYC/KYB, a Go jobs handler, a Flutter mobile app (~120 screens), a SvelteKit admin console, and a Next.js site
   • Closed a live withdrawal-fraud race condition by moving every external payout path to debit-before-pay ordering, with row-level locking and deterministic idempotency keys
   • Designed a rules-first, LLM-last fraud/AML intelligence layer: 38 deterministic detectors, LLM-drafted suspicious-activity reports, and a compliance Q&A analyst guarded by k-anonymity floors and PII redaction
   • Built an end-to-end document-generation system with large language models on Azure ML Studio, cutting manual creation time by 85% while meeting compliance requirements across three regions
   • Deployed ML models via FastAPI on Microsoft Azure, serving predictive analytics for agriculture (500+ farmers), financial inclusion (90% classification accuracy), and compliance systems
   • Automated build and deployment with per-service Azure DevOps pipelines and Bicep infrastructure, reducing time to production by 40%; quality gates include 680+ automated tests and formal VAPT remediation

Publications

Peer-reviewed

1. Saheed Faremi, Luca Longo "Integrating Convolutional Variational Autoencoders and the Gaussian Mixture Model for efficient manifold learning and clustering of spatially preserved EEG topographic maps". Brain Informatics, 2026. doi:10.1186/s40708-026-00327-9
2. Akinpelumi Saheed Faremi, Boluwaji Akinnuwesi, Elliot Mbunge, Petros M. Mashwama, Stephen Fashoto, Polite Zenzo Ncube, John Batani, Shamsudeen Ademola Sanni, Yinusa A. Faremi, Andile Metfula "Machine Learning Models for Identifying Factors Influencing and Predicting Malaria Among Children Under Five Years in Nigeria". IEEE ICTAS 2024, 2024. doi:10.1109/ICTAS59620.2024.10507142

Preprints, registered protocols & doctoral work

1. Morteza Akbari, Saheed Akinpelumi Faremi, Luca Longo "Autoencoder-Based Models for Scalp EEG: A Systematic Review of Architectures, Applications, Latent Representations, Interpretability, and Validation". OSF Registries (registered systematic-review protocol), 2026. doi:10.17605/osf.io/v2w3z
2. Saheed Faremi, Andrea Visentin, Luca Longo "Interpretable EEG Microstate Discovery via Variational Deep Embedding: A Systematic Architecture Search with Multi-Quadrant Evaluation". XAI 2026 (Late-breaking work + Doctoral Consortium track), Fortaleza, Brazil (arXiv preprint), 2026. doi:10.48550/arXiv.2605.10947 code
3. Saheed Faremi "Explainable Disentangled Representation Learning of Recurring Brain Activation Patterns via Variational Autoencoders". XAI World Conference 2025, Doctoral Proposals track, 2025.

Education

1. PhD · University College Cork

   In progress

   Neural engineering and machine learning

   Supervisor: Luca Longo

2. MSc · Technological University Dublin

   Machine learning, deep learning, data mining, visualisation, statistics

   Taught master's in computer science with a machine-learning concentration; coursework in deep learning, data mining, statistical modelling, and visualisation. Completed 2025 with a 4.0/4.0 GPA (80/100).

3. BSc · University of Eswatini

   Information technology

   Bachelor of Science in Information Technology, awarded with Second Class Honours, Upper Division (2.1).

Technical skills

AI & machine learning

PyTorch · TensorFlow · scikit-learn · CatBoost · LLMs (Azure OpenAI, OpenAI, Anthropic) · RAG · Fraud and anomaly detection · Transformer fine-tuning (BERT, RoBERTa) · Explainable AI

Programming

Python · SQL · Go · TypeScript · Bash

Cloud & MLOps

Azure (ML Studio, AI Studio, Functions, DevOps, Bicep) · CI/CD (GitHub Actions, Azure DevOps) · SLURM cluster computing

Data & systems

ETL/ELT pipelines · MySQL · Double-entry ledger systems · RESTful APIs and microservices

Statistics & analytics

Statistical hypothesis testing (Wilcoxon, ICC, multiple-comparison correction)

Recognition

1. Winner, UNESCO India-Africa Hackathon 2022 (AGRI12) 2022

   UNESCO · Ministry of Education Innovation Cell (India) · Gold medal + ₹3 lakh team prize

2. Deep Learning IndabaX: Churn Prediction Challenge 2022

   Deep Learning IndabaX · 3rd of 21

3. Google NLP Hack Series: Swahili Sentiment Analysis 2023

   Google NLP Hack Series · 4th of 29

4. UmojaHack Africa 2023: Cryptojacking Detection 2023

   UmojaHack Africa · 35th of 328 (top 11%)

Selected projects

1. AI-assisted Farmer Call Center

   Automated voice-response system letting farmers report issues via phone, SMS, or web and receive AI-routed answers. · Team engineer, Eswatini representative

   Python · Voice AI · Twilio

2. EEG microstate analysis with variational autoencoders

   Source segmentation of EEG signals via variational autoencoders, including a GMM-VAE for soft clustering. · PhD researcher

   Python · PyTorch · MNE · NumPy · scikit-learn

3. Multilingual sentiment analysis for Swahili

   Transformer models fine-tuned for sentiment classification of Swahili tweets; placed 4th of 29 in the Google NLP Hack Series. · Author

   Python · BERT · RoBERTa

Talks

1. Interpretable EEG Microstate Discovery via Variational Deep Embedding · XAI 2026 (Late-breaking work + Doctoral Consortium track), 2026
2. Machine Learning Models for Predicting Malaria in Nigerian Children Under Five · IEEE ICTAS 2024, 2024