TEMITAYO ABIONA

• Workflow Orchestration: Designed complex ML workflows with multi-step dependencies using FastAPI and LangChain, ensuring deterministic execution, retry logic, and failure isolation.
• Unified Model Serving: Integrated Vertex AI and custom model endpoints behind a common serving layer, optimising throughput and tail-latency.
• Reproducibility & Validation: Implemented structured validation of model inputs and outputs, enabling consistent evaluation.
• Data Pipelines: Built automated ETL workflows to ingest, clean, and version unstructured datasets for downstream ML usage.

• Model Benchmarking: Designed mathematically grounded benchmarks to evaluate reasoning and failure modes in large language models.
• Reproducible Evaluation: Built repeatable evaluation pipelines with clear metrics and versioned datasets.
• Cross-Functional Collaboration: Worked closely with researchers to translate evaluation insights into actionable improvements.

• End-to-End ML Pipelines: Architected reusable ML pipelines covering training, validation, deployment, and monitoring using AWS SageMaker, reducing manual intervention by 90%.
• Compute Efficiency: Redesigned cloud compute strategy using spot instances and autoscaling, achieving £100k (20%) annual cost savings.
• RAG Systems: Built production-grade Retrieval Augmented Generation pipelines integrated with Pinecone (BERTScore: 87%).
• Production Robustness: Deployed comprehensive monitoring (CloudWatch + custom metrics), reducing system downtime by 25%.

Focus: Machine Learning Pipelines, Large-Scale Data Analysis, Neural Networks.

Thesis: Generative AI for Fraud Detection in PropTech using GANs for anomaly detection.

Developed a hybrid PCA / Factor Analysis approach for predictive modelling in healthcare.