Taiwo Jegede
RAG, cloud deployment, NLP, and analytics products built with evidence, safety, and measurable outcomes.
AI Engineer and Senior Data Analyst with a Master’s in Artificial Intelligence and 4+ years of experience in architecting serverless AI products, performance monitoring, and scalable data pipelines. Specializes in translating complex unstructured behavioral data into rigorous statistical models and actionable strategies. Expert in Python, SQL, NLP (BERT, Transformers), Power BI, and AWS (Bedrock, Lambda, Glue), with a proven track record of leading technical initiatives that drove a 23% ROI increase and reduced data retrieval latency by 50%.
GPA: 4.0/4.0
GPA: 4.53/5.0
Bilingual RAG agricultural advisory platform for Arkansas farmers, grounding rice, soybean, poultry, and spray-compliance guidance in Extension documents, live soil data, weather context, and citation safeguards.
AI-assisted IEP documentation product with RAG-grounded drafting, deterministic federal and state validation, local PII redaction, audit logging, human review gates, and .docx export for special-education teams.
An intelligent AI agent designed for supply chain negotiations using strategic algorithms and real-time simulation.
Cloud-native negotiation platform migrated to AWS Lambda + Amplify with a scalable FastAPI RESTful backend.
Comprehensive suite including VADER Sentiment Analysis, NER tagging, Spam Detection, and Topic Modeling implementations.
Statistical analysis projects including Titanic Survival Prediction (97.2% acc) and interactive Power BI dashboards.
End-to-end ML pipeline that identifies at-risk telecom customers, ranks them by churn probability (0.808 recall, 0.848 AUC-ROC), and explains predictions via an interactive Streamlit dashboard.
Three AI modules tackling the 2025 higher-ed demographic cliff: SARIMA/Prophet forecasting, XGBoost lead and yield prediction with bias audit, and a RAG admissions chatbot with 100% trick-refusal rate.
Identifies at-risk students 6-8 weeks before formal academic warnings using behavioral signals from 32K OULAD records. Calibrated XGBoost achieves 0.844 AUC-ROC with SHAP-explained advisor dashboard.
AI triage tool for campus counseling centers: behavioral risk classifier, NLP distress detector with LIME explainability, and a crisis-aware chatbot with mandatory override layer. Advisor aid, not clinical diagnosis.