Opeyemi Bamigbade
Profile

Opeyemi Bamigbade

AI Research Engineer | ML Systems | PhD Candidate

About Me

I build model-centric ML systems for multimodal representation learning, spanning model design, training, evaluation, and inference. My work sits at the boundary between research and deployment, with a focus on translating experimental ideas into robust, production-aware systems that operate efficiently at scale.

AI Research & Model Development

  • Multimodal and representation learning for vision and vision–language models
  • Algorithm design, experimentation, and ablation-driven evaluation
  • Modular model components designed for reuse and extensibility
  • Fine-tuning, adaptation, and analysis of model behavior and failure modes

Model-Centric ML Systems

  • End-to-end training, evaluation, and inference pipelines for ML models
  • Inference-first system design with emphasis on efficiency and reliability
  • Performance optimization across compute, memory, and data flow
  • Packaging and deployment of research models for real-world integration

Experience

Senior AI Engineer

Curacel

Dec 2021–2023On-site

Led the design and deployment of company’s first production AI system for automobile insurance claims automation. Built and productionized DProcessor, an OCR-driven data processing pipeline that increased claims throughput by 91%, and deployed deep learning–based object detection models as scalable services.
Owned the end-to-end lifecycle of multiple ML systems, spanning model development, evaluation, inference optimization, and cloud deployment. Designed deployment-ready architectures using AWS services, packaged models as reusable microservices, and integrated vector search capabilities to support downstream decision systems. Mentored engineers on production-aware ML practices and system-level performance trade-offs.

Machine Learning & Backend Engineer

Philanthrolab

Dec 2021–2023Remote

Designed and implemented ML-driven personalization systems for social and human services platforms. Developed NLP-based models for eligibility scoring and recommendation, translating research prototypes into deployable services used in production workflows.
Built and maintained the supporting data and backend systems required for model training, evaluation, and inference, including user behavior tracking, data pipelines, and API-driven integrations. Contributed across the stack to ensure reliable interaction between ML models and application services.

Machine learning Engineer

Clinify

May 2020–2021Remote

Designed and deployed machine learning systems for healthcare applications, spanning model development, system integration, and evaluation. Worked closely with product teams to translate research ideas into reliable, production-ready ML components.
Contributed to system-level testing strategies for complex ML-backed applications and supported backend integration of deployed models through API-based services.

Machine learning Engineer

Data Science Nigeria

June 2019 - 2021Remote

Worked on a range of applied ML and NLP systems, including ranking models, conversational agents, and computer vision pipelines. Designed and deployed resource-efficient models, including edge-deployed solutions on constrained hardware (Raspberry Pi), operating under strict compute and memory constraints.
Built modular ML services for client integration and explored system trade-offs across performance, accuracy, and deployment constraints in real-world settings.

Education

PhD in Computer Science (in progress)

South East Technology University

2022 – Present Ireland

Research centered on multimodal computer vision systems, with a focus on representation learning, model robustness, and system-level evaluation. My work explores colour-augmented embeddings for indoor visual understanding, attention-driven segmentation for anomaly localization, and open-set detection under real-world constraints. I treat applied domains as testbeds for designing generalizable, scalable vision and multimodal learning methods, with emphasis on efficient training and inference pipelines.

B.Sc. Systems Engineering (First-Class Honours)

University of Lagos

2014 – 2018 Nigeria

Focused on system design and applied machine learning. Final-year project involved developing a machine learning–driven SaaS platform for automated Gleason score prediction, exploring end-to-end model deployment for healthcare decision support.

  • President, University of Lagos Artificial Intelligence Club
  • Ambassador, Zindi Africa (Data Science Competition Platform equivalent to Kaggle)

Skills

Programming & Frameworks

  • Python | Go
  • PyTorch | TensorFlow | CUDA
  • Scikit-learn | OpenCV | NLTK

Modeling & Learning Paradigms

  • Multimodal & Vision-Language Models (VLMs)
  • Representation Learning | Contrastive Learning
  • Attention Mechanisms, Transformers
  • Self-Supervised | Few-Shot Learning
  • Diffusion Models

Model-Centric ML Systems

  • Tensor Operations | Sparse Matrices | Quantization
  • Inference optimization: ONNX | TensorRT | TVM
  • Training & scaling: Hugging Face | OpenCLIP | DeepSpeed

Systems, Deployment & Tooling

  • MCP | FastAPI | Flask
  • ONNX | TensorRT | TVM
  • AWS | GCP
  • Git | CI/CD
  • MLOps practices and pipelines

Projects

VAAS: Inference-First Vision Anomaly Scoring Library

Open-Source Maintainer

VAAS is an inference-first, research-driven vision library for attention-based anomaly scoring. It operationalizes Vision Transformer attention patterns and patch-level self-consistency to produce fine-grained anomaly maps, enabling localized inspection of visual irregularities without requiring task-specific retraining.
The library is designed for deployment and reuse, providing a lightweight inference interface that loads pretrained models from Hugging Face and integrates cleanly into downstream analysis pipelines. VAAS emphasizes interpretability, modularity, and system-level reliability, bridging published research with practical ML system integration.

Vision Transformers Attention Mechanisms Patch-level Analysis Inference-First Design Model Interpretability Open-Source ML Systems

OpsMate: Agent-Based AI Assistant System

Project Lead

OpsMate is an AI agent–based assistant system implemented as a Chrome extension, designed to support context-aware, persistent interactions through an augmented LLM backend. The system combines a lightweight client interface with an MCP-backed server architecture to manage conversation state, tool access, and user-specific context.
The project emphasizes system integration and reliability, including login-based persistence, structured Markdown rendering, and a responsive UI layer. OpsMate serves as a platform for exploring agent orchestration, context management, and LLM-backed system design in real-world user-facing

Agent Systems Augmented LLMs MCP Server Architecture Context Management Chrome Extensions AI System Integration

ML in Containers: Model Deployment & CI/CD Reference Project

A reference implementation for training, containerizing, and deploying a vision model in production-like environments. The project demonstrates model serving with TensorFlow Serving, API-based inference, and a complete CI/CD workflow using GitHub Actions for automated build and deployment.
Designed as a systems consolidation project, focusing on reproducibility, deployment correctness, and industry-standard practices for ML model lifecycle management.

Model Serving Containerized ML CI/CD for ML Systems TensorFlow Serving Deployment Workflows Docker Swarm

Danfojs: A JavaScript-based Data Analysis Library

Open-Source Contributor

Contributed to JavaScript-based machine learning and data analysis libraries, with a focus on documentation, usage examples, and practical integration patterns for TensorFlow.js. Work supported clearer adoption of tensor-based data manipulation and model inference in browser and Node.js environments.

Tensors DataFrames JavaScript ML TensorFlow.js Ecosystem

Pylette: A Python Library for Image Colour Extracting

Open-Source Contributor

Contributed to a lightweight Python library for image colour analysis, supporting extraction and analysis of dominant colour palettes from visual data. Work focused on improving usability and feature support for colour-based representation and image processing workflows.

Image Processing Colour Analysis Colour Spaces Python

Decision Support System

Developed a decision support system for assisting pathologists in prostate cancer assessment using deep learning–based analysis of histopathological images. The system integrates trained vision models into an interactive interface to support diagnostic workflows and exploratory analysis. The project demonstrates early experience in applied computer vision systems, model integration, and user-facing ML tools for domain-specific decision support.

Medical Imaging Computer Vision Decision Support Systems Model Integration

ML Model Deployment & Data Pipeline

Built a custom data pipeline and deployment workflow for a predictive machine learning model, covering data ingestion, model serving, and API-based inference. The project focuses on demystifying end-to-end ML system construction using reproducible, production-style patterns. Serves as a reference implementation for integrating data pipelines, trained models, and deployment interfaces in applied ML systems.

Data Pipelines Model Deployment API-Based Inference ML System Integration

Research Publications

VAAS: Vision-Attention Anomaly Scoring for Image Manipulation Detection in Digital Forensics

Introduces Vision-Attention Anomaly Scoring (VAAS), a dual-module framework that combines global attention-based anomaly estimation using Vision Transformers with patch-level self-consistency scoring derived from segmentation embeddings. The method enables fine-grained localization of visual irregularities and supports interpretable anomaly analysis in complex image settings.

Improving Image Embeddings with Colour Features in Indoor Scene Geolocation

Proposes a colour-aware representation learning architecture that integrates dominant colours and multi-space colour histograms with visual embeddings. The work demonstrates consistent improvements in indoor scene understanding, highlighting the impact of explicit colour representations and comparative advantages of classification-based approaches over metric learning in colour-augmented settings.

Computer Vision for Multimedia Geolocation: A Systematic Literature Review

Presents a systematic review of 123 studies on computer vision and AI methods for multimedia geolocation. The review analyzes techniques, datasets, and evaluation strategies, highlighting applications in digital forensics and human trafficking investigations, and outlines open challenges and future research directions for geolocation-based evidence gathering.

Peer Reviewing

Serve as a peer reviewer for AI, Machine Learning, and Computer Vision venues, with a focus on evaluating deep learning architectures, multimodal and colour-aware representations, and system-level rigor. Reviews emphasize technical soundness, reproducibility, and deployability of proposed methods.

Consultations

Tegus

Provide expert advisory support to startups on vector search systems and similarity-based retrieval, including algorithmic design and practical trade-offs across modern vector databases.

Vector Search Similarity Retrieval Embedding Systems Vector Databases

MLOps Community

Author a technical writing series on Test-Driven Development for MLOps, combining conceptual guidance with practical code examples to support reliable and reproducible ML system development.

Datakirk

Delivered expert-led training sessions on data science and machine learning for practitioners and technical audiences, covering applied concepts, model development workflows, and best practices for real-world ML systems.

Blogs

Building Production Machine Learning Systems

How to deploy scalable, production-grade ML systems on Google Cloud with modern MLOps workflows.

Test-Driven Development in MLOps

How to apply TDD to ML workflows to improve reliability, catch data issues early, and build production-ready pipelines.

Connect

Email: bamigbadeopeyemi@gmail.com

LinkedIn: Opeyemi Bamigbade

Twitter: opeyemibami

Github: opeyemibami

Scholar: Opeyemi Bamigbade

Medium: Opeyemi Bamigbade

🗄️ Archive

A collection of past projects, certifications, experiments, and legacy work.