Opeyemi Bamigbade
Profile

Opeyemi Bamigbade

AI/ML Systems Engineer

PhD Candidate in Multimodal AI

📍 Ireland · Open to AI Systems & Platform Engineering Roles

About Me

AI/ML Systems Engineer building at the intersection of research and production systems. My work has taken me from deploying machine learning systems in industry to exploring the field through research. I now focus on inference systems, runtime operations, and production infrastructure.

What I Build

ML Systems

  • Builds AI inference systems across serving and orchestration workflow.
  • Develops observable ML services optimised for production workloads
  • Designs runtime components for scalable model execution.
  • Focus : Runtime Systems · Inference Engineering · Model Serving

Applied AI Research

  • Translates research into deployable AI capabilities.
  • Evaluates and improves model behaviour under constraints.
  • Connects research insights with production AI systems.
  • Focus: Inference Reliability · Runtime Observability

Projects

Platframe operational runtime platform
Available to Demo Operational Platform Runtime Coordination Replay Systems ModelOps

Platframe: Operational AI Platform for Bounded AI Workflows

Governed Workflow Execution · Replayable Runtime Systems

Problem: AI-native operational platforms, such as insurance claims management systems, become unreliable when workflow execution and runtime behaviour are non-deterministic. Operational decision paths also become difficult to govern across multi-step workflows.
Built System: An operational AI platform for governed workflows, replayable runtime history, and bounded operational execution.
Subsystems:
  • OpsMate: Stateful coordination agent for workflow assistance and execution support.
  • Infera: Inference coordination layer for provider routing, execution control, and evaluation.
  • VAAS: Vision analysis system for evidence integrity scoring, localisation, and anomaly analysis.
  • Governance: Governance layer for workflow validation, escalation, and review control.
Operational Focus: Designed around runtime continuity, governed workflow progression, replay reconstruction, bounded execution behaviour, and operational lifecycle reliability.
4 Bounded Runtime subsystems
3 Human-in-the-Loop Components
1 Correlated Operational replay
Notes
Preview Live View Code All Notes
VAAS research and inference runtime system
R&D · Research to Production Inference Systems Vision Transformers Interpretability

VAAS: Research to Inference Runtime System

Vision-Attention Anomaly Scoring · Published at DFRWS EU 2026

Research Problem: Existing image manipulation detectors rely heavily on pixel-level artefacts and binary decisions. They often miss semantic manipulations and provide limited spatial reasoning around anomaly intensity and localisation.
Built Model Architecture: Dual-module framework combining a Vision Transformer global attention module (Fx) with a Segmentation patch-consistency module (Px), fused via cross-attention to produce continuous hybrid anomaly scoring (S_H) and dense spatial anomaly maps.
Inference:
  • Published fast inference library with documented APIs for operational workflows and integration.
  • Multiple versioned model variants hosted on Hugging Face and loadable through standard pretrained runtime interfaces.
  • Exposes scoring, heatmap generation, and mask extraction APIs for extensible downstream use across vision tasks.
  • CPU and GPU inference support with container-ready deployment and reproducible notebook workflows.
Operational Focus: Designed around explainable anomaly reasoning, reusable inference workflows, operational deployment readiness, and interpretable visual integrity analysis.
100+ Monthly Downloads
5+ Model Versions
26+ GitHub Stars
HF Hugging Face Availability
Research Outputs & Notes
View Code Model on HF PyPI Paper

Experience

AI Research Engineer / PhD Researcher

South East Technological University (SETU)

2022–Present Ireland

  • Designed multimodal and attention-based AI systems focused on robust perception, anomaly localisation, and evaluation reliability across computer vision tasks.
  • Built end-to-end PyTorch training and inference workflows covering reproducible experimentation, model evaluation, and performance analysis across research systems.
  • Developed modular AI libraries and inference APIs with versioned Hugging Face releases for reproducible evaluation and downstream integration.
  • Optimised model training and inference workflows through profiling of GPU memory usage, throughput, and execution bottlenecks.
  • Published peer-reviewed research in multimodal AI and computer vision, with deployment-focused implementations released as open-source software.
3+ Peer-reviewed Publications
5+ Released Model Variants
5+ Conferences and Presentations
27+ Citations

AI Engineer

Curacel

Dec 2021–2023Hybrid

  • In a team of 5, led the design and delivery of insurance claim automation, applying computer vision techniques including object detection, damage localization, and image-based classification to automate assessment workflows and downstream decision systems.
  • Built cloud-native MLOps workflows from model training to inference deployment across AWS and GCP resources with continuous integration and delivery.
  • Built and operationalised an OCR-driven medical claims processing pipeline for resolving unstructured pharmaceutical terminologies using vector database retrieval across health insurance workflows, improving claims throughput by 91% while increasing processing reliability and operational stability.
  • Contributed to deployment architecture, inference performance optimisation, and systems engineering decisions for production AI workloads.
2+ owned systems
91% Claims Throughput Improvement

AI & Backend Engineer

Philanthrolab

Dec 2021–2022Remote

  • Designed and deployed ML-driven personalization systems for social and human services platforms, delivering NLP-based eligibility scoring and recommendation models integrated into live application workflows.
  • Built backend services supporting model training, evaluation, and inference, implementing data pipelines, user behavior tracking systems, and API-driven microservices connecting ML systems to production applications.
  • Optimised system performance, data flow, and operational reliability across the ML lifecycle to support stable production deployment and scalable inference workloads.
1+ owned systems
2+ R&D

Machine learning Engineer

Clinify

May 2020–2021Remote

  • Designed and deployed machine learning systems for healthcare applications, spanning model development, evaluation, and integration within production environments.
  • Collaborated with product and engineering teams to translate research-driven concepts into reliable, deployable ML components integrated into application workflows.
  • Contributed to system-level validation of ML-backed services and supported backend integration through API-based systems, ensuring reliability across data pipelines and production workflows

Machine learning Engineer

Data Science Nigeria

June 2019 - 2020On-site

  • Developed applied ML and NLP systems including topic modelling and ranking models, conversational chatbots, and computer vision pipelines for AI applications.
  • Designed and deployed resource-efficient models for constrained environments, including edge deployments on hardware such as Raspberry Pi under strict compute and memory limits.
  • Built modular ML services for client integration and evaluated system trade-offs across performance, accuracy, and deployment constraints in production environments.

Writings

Platframe: Building an AI-Assisted Investigation Platform
System Note AI Platform Operational Systems

Platframe: Building an AI-Assisted Investigation Platform

A walkthrough of the environment, architecture, workflow, subsystem boundaries, and lessons from building Platframe, a multi-stage AI workflow system using insurance investigations as the operational environment.

Article View Code
Agent Learning
Research Note Agentic Workflow LLM Systems

Agent Learning: Execution-Aware Policy Optimization for LLM Tool Systems

Tool orchestration in LLM agent systems as a learnable execution policy, optimising for cost, latency, and constraint satisfaction rather than static heuristic control.

Article PDF View Code

Research Publications

VAAS paper

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

FSI DFRWS EU March 2026

Introduces Vision-Attention Anomaly Scoring (VAAS), a dual-module framework combining global attention-based anomaly estimation in Vision Transformers with patch-level self-consistency scoring for fine-grained anomaly localization.
Formalizes attention-driven representation analysis as a structured anomaly detection mechanism, advancing modular and scalable approaches to interpretable vision-based anomaly modeling.

Paper PDF Slides
CEmbed paper

Improving Image Embeddings with Colour Features in Indoor Scene Geolocation

IEEE ACCESS April 2025

Introduces a colour-aware representation learning architecture integrating dominant colour descriptors and multi-space colour statistics with learned visual embeddings to enhance indoor scene understanding.
Establishes explicit colour modeling as a complementary signal to learned representations and analyzes the behavior of classification and metric learning objectives under colour-augmented settings, informing architectural and training design decisions.

Paper PDF Slides
Systematic Review

Computer Vision for Multimedia Geolocation: A Systematic Literature Review

arXiv Feb 2024

Synthesizes 123 studies on computer vision and AI approaches to multimedia geolocation, examining architectural trends, dataset design, evaluation protocols, and application contexts across visual geolocation systems.
Identifies methodological gaps, recurring limitations, and emerging design patterns, outlining open challenges and future directions with emphasis on robustness, generalization, and real-world deployment constraints.

Paper PDF

Skills

Core Languages & ML Frameworks

  • Python
  • PyTorch · TensorFlow
  • vLLM · LiteLLM · DeepSpeed

Multimodal AI

  • Multimodal & Vision–Language Systems
  • Attention Mechanisms & Transformers
  • Representation Learning
  • Parameter-Efficient Fine-Tuning

Inference & Runtime Systems

  • Model Serving & Inference Workflows
  • Runtime Orchestration & Execution Systems
  • Evaluation & Observability Pipelines
  • GPU-Backed Inference Infrastructure

Backend & Infrastructure

  • FastAPI · MCP
  • Containerisation & Reproducible Pipelines
  • CI/CD & Deployment Workflows
  • Cloud-Native ML Infrastructure

Education

PhD in Vision and Multimodal AI (Expected 2026)

South East Technology University

2022 – Present Ireland

Develops foundational expertise in multimodal AI systems, spanning representation learning, robustness, and scalable model design. Research focuses on controllable representation learning and system-level evaluation for real-world visual reasoning, including anomaly detection and scene understanding.

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

University of Lagos

2014 – 2018 Nigeria

Developed early foundations in systems design and applied machine learning.
Final-year project: Built a machine learning–driven SaaS platform for automated Gleason score prediction, covering model development, validation, and deployment within a healthcare decision-support system.

Consulting

Tegus - Advisory Support

Vector Search Indexing Strategies

Advisory support to startups on vector search architectures and similarity-based retrieval systems, covering embeddings strategies, indexing, and trade-offs across vector DB

7+ consultations
Learn More

Peer Reviewing

AI Systems Computer Vision Multimodal AI

Reviews research submissions for leading AI and computer vision conferences and journals, with focus on multimodal systems, inference reliability, and evaluation methodologies.

13+ Reviews
Learn More

MLOps Community - Technical Author

TDD Reproducible Pipelines Model Lifecycle Testing

Authors a technical series on Test-Driven Development for ML systems, outlining structured approaches to reliability, reproducibility, and lifecycle validation. Combines conceptual guidance with implementation-level examples.

3+ Writings
Learn More

Datakirk - Technical Instructor

Applied Machine Learning Model Development Workflows

Delivered expert-led training on data science and machine learning for technical practitioners. Covered model development workflows, applied learning paradigms, and engineering best practices for reliable and production-aware ML systems.

2+ Lectures
Learn More

🗄️ Archive

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

Connect

Email: bamigbadeopeyemi@gmail.com

LinkedIn: Opeyemi Bamigbade

Twitter: opeyemibami

Github: opeyemibami

Scholar: Opeyemi Bamigbade

Medium: Opeyemi Bamigbade