AI & Data Engineer at KAREVAULT (UPSCALEMS)

Enabled project predictive schedule analytics by building ETL pipelines to transform Primavera P6 and MS Project data into structured, queryable formats supporting cross-project analysis. Accelerated AI delay-analysis processing by 4× using Redis caching and Celery-based asynchronous task orchestration for large-scale project workloads. Improved LLM response reliability by implementing automated guardrails and LLM-as-a-Judge evaluation pipelines using MLflow. Monitored token usage, latency, and evaluated model performance/coherence using MLflow. Increased agent response accuracy and iteration speed by replacing static prompts with DSPy-optimized, self-improving modules across agent workflows. Improved schedule validation accuracy by developing multimodal reasoning pipelines using ReAct & CoT, integrating tabular project data with site imagery. Built scalable property listing ingestion pipelines using web scraping, Kafka queues, and Elasticsearch indexing for high-performance search an

ML Engineer at PREDICTIVE-EDGE

Reduced data preparation time across retail stores by designing scalable PySpark ETL pipelines to unify multi-source transactional data into centralized MongoDB analytics storage. Enabled data-driven procurement decisions by developing time-aware XGBoost forecasting models achieving 89% revenue and inventory prediction accuracy. Increased model performance by over 30% through Optuna-driven Bayesian hyperparameter tuning. Standardized 50+ inconsistent retail data features via robust preprocessing and transformation pipelines using Pandas and PySpark. Reduced deployment cycle time from days to hours by managing experiment tracking, versioning, and automated production deployment through MLflow. Maintained forecasting reliability by deploying real-time pipelines to detect model and data drift and automatically trigger retraining workflows.

AI & Data Engineer at SLIMPREP (BAI-Innovations)

Streamlined energy analytics ingestion by designing Airflow-driven ETL pipelines integrating VRM APIs into Azure Data Lake and Azure SQL systems. Enabled scalable data processing services by deploying PySpark and Django-based transformation pipelines on Azure App Services. Delivered production-ready APIs using Django REST Framework hosted on Azure infrastructure. Improved energy consumption forecasting accuracy through ensemble ML models trained on Azure SQL analytics data. Enabled contextual dashboard querying by building multi-agent RAG chatbots using FastAPI and LangGraph. Increased LLM safety and coherence by integrating automated evaluation and guardrail pipelines with MLflow. Reengineered legacy PHP systems into modern Python codebases, accelerating debugging and improving scalability. Accelerated production releases by implementing CI/CD workflows and automated testing pipelines in Azure environments. Reduced deployment friction by containerizing ML models using Docker and manag

Machine Learning Ops Engineer at TERRAHAPTIX

Enabled real-time drone vision capabilities by designing preprocessing pipelines supporting CNN and Vision Transformer workloads. Reduced training cost while preserving model performance by fine-tuning large foundation models using LoRA and QLoRA techniques. Enabled autonomous drone localization through SLAM architectures using stereo video inputs and HD mapping. Improved real-time drone data transmission reliability via Kafka-driven event pipelines and microservice infrastructure. Enhanced training datasets by collaborating with research team on GAN-based aerial frame synthesis and augmentation pipelines. Reduced inference latency by deploying optimized model serving pipelines via vLLM and quantization techniques. Accelerated large-model training convergence through distributed multi-GPU scaling with DeepSpeed parallelization strategies.

Computer Vision Research Assistant at ODUDUWA UNIVERSITY, Osun State

Enabled near real-time video processing by applying pruning and quantization techniques to reduce model computational overhead. Contributed to autonomous driving research by developing transformer-based Birds-eye view reconstruction architecture for stereo vision scene reconstruction. Increased department enrollment by assisting teaching efforts in photogrammetry and 3D computer vision. Improved training efficiency by optimizing large-scale video preprocessing pipelines for multi-architecture compatibility. Guided research direction by benchmarking 3D-CNN, Spatial-RTNs, and Transformer models for temporal video understanding.