F2-CUSTOMER-RAG-AGENT · 2026
Customer-interaction RAG agent
A confidence-gated customer-operations agent that combines policy citations, scoped synthetic ERP lookups, prompt-injection handling, and inspectable human escalation.
Read case studyApplied AI · GenAI · responsible systems
Production-minded AI systems with inspectable evidence.
I build decision-support workflows that connect thoughtful problem framing, credible baselines, evaluation, and the operational details that make an AI system trustworthy.
01
Selected work
Evidence-led projects across predictive ML, retrieval, agents, and operational evaluation.
F1-FALL-RISK-PREDICTION · 2026
Seven-day fall-risk prediction
A reproducible fall-risk prioritisation workflow for fictional aged-care residents, connecting calibrated baselines, review capacity, human explanations, failure analysis, and simulated monitoring.
Read case study
F6-MENTAL-HEALTH-READMISSION · 2026
Mental-health inpatient readmission prediction
A leakage-safe discharge-readmission workflow for a fully synthetic Australian acute-care cohort, focused on care transitions, calibration, review capacity, subgroup behavior, and documentation drift.
Read case studySynthetic / simulated evidence All outcomes are synthetic or simulated portfolio evidence.
02
What I build around the model
The useful part is the surrounding system: evaluation, human boundaries, observability, and honest communication.
Predictive ML
Baselines, calibration, slices, and capacity-aware review.
RAG + agents
Retrieval, tools, guardrails, citations, and escalation.
Evaluation
Metrics that connect model behavior to the actual decision.
Monitoring
Drift, missingness, dependency failures, and response.
Responsible AI
Clear boundaries, synthetic data, and inspectable limits.
03
A short introduction
A practical, research-minded approach to applied AI.
I am an applied AI builder working across predictive machine learning, retrieval-augmented generation, and evaluation-led product thinking. My portfolio focuses on systems that are useful because their boundaries are visible.
Across each project, I make the decision, evidence, assumptions, failure modes, and human handoff visible enough to inspect without running the entire system.
Open to thoughtful problems
Let’s compare notes.
Add an email address in app/profile.ts