Kesser Karim

Machine Learning Engineer with 3+ years building production-ready AI systems in healthcare and applied research. Currently completing PhD (University of Portsmouth) on fairness-aware Type 2 Diabetes risk prediction for underrepresented populations. Azure-certified (DP-100) with deployed models serving live predictions via FastAPI + Docker. Proven track record designing and shipping end-to-end ML pipelines from data acquisition to Fairlearn bias audits to Hugging Face Spaces deployment. Seeking ML Engineer / Applied Scientist roles in healthcare AI, EdTech, or fairness-focused teams in UAE or remote UK.

• Built end-to-end fairness-aware ML pipeline for T2D risk prediction targeting South Asian populations (2–4× higher incidence), achieving 0.847 AUROC with <0.05 subgroup disparity after threshold optimisation
• Reduced demographic parity gap by 67% (0.18 → 0.06) using Fairlearn MetricFrame + threshold optimisation; full audit in public Jupyter notebooks with calibration curves per ethnicity/sex/age band
• Deployed production API (FastAPI + Docker) serving real-time risk scores + SHAP explanations; CI/CD via GitHub Actions with pytest coverage >85%; live demo on Hugging Face Spaces

• Designed ML system identifying at-risk students across 1,200-student cohort, flagging 18% for targeted intervention with 0.72 precision
• Automated monthly batch predictions via AWS Lambda + S3, reducing manual triage by ~15 hrs/month; Streamlit dashboard onboarded 8 non-technical counsellors

• Building RAG system over NICE NG28 + ADA clinical guidelines; 400+ pages chunked into vector embeddings; hybrid dense + BM25 retrieval via LangChain; targeting <2 s latency with citation transparency

• Thesis: Fairness-Aware Machine Learning for Type 2 Diabetes Risk Stratification in South Asian Populations
• Investigating algorithmic bias in clinical ML trained on predominantly white European datasets; publishing on calibration fairness and subgroup-specific risk thresholds
• Collaborated with NHS Connected Bradford on de-identified EHR analysis; privacy-preserving pipelines for multi-ethnic cohort studies
• Publications: 2 conference papers submitted (FAccT 2026, ML4H 2026); workshop presentation at AI UK 2025
• Supervisor: Dr. Rinat Khusainov

Final Project: Fairness-Aware Feature Engineering for Diabetes Prediction (92/100)

• Deployed predictive models for student performance tracking (scikit-learn + PostgreSQL); automated weekly risk reports reducing manual analysis by 70%
• Built data pipelines ingesting 12K+ student records across Google Classroom, MobyMax, Zearn; ETL workflows with Make.com + Python for real-time dashboards
• Led A/B testing on adaptive learning interventions (N=300 students); measured 12% improvement in math proficiency via Bayesian analysis of pre/post assessments
• Mentored 6 teachers on data-driven instruction; trained non-technical staff to interpret ML-flagged student cohorts

• Taught Cambridge IGCSE/A-Level Computer Science (Python, algorithms, data structures) to 200+ students; 95% pass rate, 40% achieving A*/A grades
• Designed curriculum integrating ML concepts and AI ethics into secondary education; piloted AI ethics module with 80 Year 11 students

References available upon request  —  kesserkarim.com