Selected Case Studies

Selected Case Studies

Back Home

Case studies across medical AI risk-warning prototypes, healthcare and public-health dashboards, enterprise healthcare technology delivery, and 0-to-1 product and workflow operations.

Case 01

Oncology High-Risk Prescribing Warning System Prototype

Medical AI / Digital Health / Explainable Alerts / MSc Independent Research
Context

MSc independent research exploring how high-risk prescribing patterns in oncology — particularly psychotropic medicines and CNS depressant combinations — could be surfaced for pharmacist or clinician review. The aim is a review-support prototype, not a treatment recommendation engine.

Challenge

Real-world prescribing data is dense and time-ordered: the same medication combination can be routine in one window and high-risk in another. The challenge was to design candidate trigger logic and explainable outputs that a reviewer can trust, while staying strictly within a non-clinical, proof-of-concept boundary.

What I Did
  • Conducted literature and data-feasibility research mapping candidate oncology cohorts, medication timelines, drug classes, observation windows, and proxy safety events (emergency visits, admissions, falls, delirium).
  • Designed candidate trigger logic around drug-class overlap, medication burden, and age / prior-event explanation context.
  • Built explainable yellow/red review-priority alerts that output triggered classes, time window, rationale fields, and a review prompt.
  • Planned a review-interface structure: medication timeline, active alerts, trigger rationale, patient stratification, review queue, and candidate-logic notes.
Relevant Skills
Real-world data mappingRisk flag designExplainable alertsClinical workflow mappingPythonMedical AI
Key Outputs

Real-world data feasibility

Mapped cohort identification, inclusion/exclusion, observation windows, drug-class grouping, and proxy safety events using real-world healthcare data structures. Aggregate-level feasibility exploration only.

Candidate warning logic

Examples include opioid + sedating-medication overlap within one window, psychotropic/sedating medication burden, and age / prior falls-or-delirium explanation context — each paired with an explanation output.

Explainable alert output

A prototype alert card: review-priority level, triggered medication-class combination, time window, explanation fields, and a prompt to clinical or pharmacy review.

Boundary & data ethics

Non-clinical proof of concept. Yellow/red indicate prototype review priority only, not validated clinical risk stratification. Candidate logic requires supervisor, clinical, or pharmacy validation and does not replace clinical judgement. No patient-level or restricted data is shown.

Business / Operational Value

Demonstrates how real-world healthcare data, candidate rules, and explainable outputs can be combined into a review-support prototype — connecting data science, healthcare context, and product thinking. It is a non-clinical proof of concept that supports, and does not replace, clinical review.

Case 02

Women's Health Access Pressure Dashboard

Digital Health / Public-Health Data / Dashboard / Decision Support
Context

Designed for a UK women's-health startup-style regional research and strategy scenario: which regions in England should a small team investigate first for access pressure, and why? It uses public aggregate data only.

Challenge

Different signals use different units and tell different stories — long waits, deprivation, and population scale don't move together. The challenge was to combine them transparently without implying a clinical or individual-level risk judgement.

What I Did
  • Combined three public aggregate signals: England gynaecology waiting-time statistics, the English deprivation index, and England female population estimates.
  • Normalized waiting pressure, access friction, and affected-population scale to 0–100 indicators and combined them through an adjustable weighted priority index.
  • Designed dashboard storytelling: regional ranking, driver decomposition, a map view, and a selected-region explanation card.
  • Documented method and limitations — awareness, trust, transport/time-off, and care-quality barriers lack comparable public data and remain follow-up questions.
Relevant Skills
Public-health indicatorsExplainable dashboardsWeighted indexingGeo analysisDashboard storytellingPython / Plotly
Key Outputs

Three signals & sources

Waiting pressure (gynaecology waiting times, Mar 2026), deprivation & access friction (English deprivation index, 2025), and affected-population scale (female population estimates, mid-2024).

Weighted priority index

Each signal normalized to 0–100 and combined through adjustable weights (default 50 waiting / 30 deprivation / 20 population), so users can test scenarios rather than trust a single fixed score.

Dashboard storytelling

Regional ranking with a regional-mean line, driver decomposition by signal, an England map (warmer = higher pressure), and a selected-region 'why this region' card with a next-investigation hypothesis.

Boundary & data scope

Public aggregate data only — no patient-level, user-level, employer-level, or internal provider data. The priority index supports regional exploration and decision support; it is not a clinical risk score and does not replace clinical judgement.

Business / Operational Value

Turns scattered public datasets into a transparent, adjustable decision-support tool that shows not just which regions rank highest but why — positioning the index as a discovery tool for strategy, explicitly not a clinical risk score.

Case 03

Johnson & Johnson Healthcare Technology Delivery Case

Healthcare Technology / Cloud Access / Security & Compliance / Cross-functional Delivery
Context

In a healthcare technology environment at Johnson & Johnson, China-region business modules needed cloud access while accounting for data isolation, security compliance, network access, and global deployment constraints. Business users needed a clear onboarding path; technical and global teams needed early visibility of China-specific risks and dependencies.

Challenge

The global Azure environment and the China-region environment were not always aligned — services, permissions, and access patterns available globally were not always available in China. Selected components had to be validated before global teams could configure networking and access.

What I Did
  • Acted as a translator between business stakeholders and technical teams, converting network architecture, access-control logic, and security dependencies into business-actionable rollout steps.
  • Coordinated business, security, network, Microsoft, and global stakeholders to support delivery.
  • Ran development-environment proof-of-concept checks and produced risk lists and validation feedback to reduce rollout rework.
  • Created SOPs and user-facing documentation for cloud resource requests, access control, network issues, and platform usage.
Relevant Skills
Microsoft AzureCloud onboardingAccess controlSecurity & complianceStakeholder coordinationTechnical documentation
Key Outputs

Localised cloud access architecture

Mapped China-region module onboarding into compliant Azure access architecture and validation documentation, accounting for data isolation, security, and network dependencies before global handover.

Dev-environment PoC & risk list

Conducted development-environment proof-of-concept checks and produced risk lists, validation feedback, architecture notes, and cross-team handover materials.

Cloud cost & optimisation reporting

Analysed Azure usage, cost, risks, and optimisation actions through monthly Excel/PPT reporting, contributing to a 16% reduction in average monthly cloud costs.

Reusable SOPs & handover

Created SOPs and user-facing documentation for cloud resource requests, access control, network issues, and platform usage to reduce repeated questions and rework.

Business / Operational Value

Shows I understand the full journey of a technical solution inside a healthcare enterprise — from requirements clarification, access security, and risk identification through to business delivery. Healthcare data products depend on controlled data environments, clear permission boundaries, and security-compliant workflows, not only models or interfaces.

Case 04

0-to-1 Product & Customer Workflow Operations

0-to-1 Product / CRM / Customer Journey / A/B Testing / Community Ops
Context

A new customer-facing business (Tada Coffee & Bistro) needed a full product and operations loop from 0 to 1: positioning, customer journey, CRM, service workflow, and online/offline operations.

Customer Flow

Community engagement -> WeChat mini-program -> ordering / booking / payment -> service workflow -> feedback & CRM follow-up -> product iteration

Business / Operational Value

Directly relevant to digital health: the same building blocks — user journey design, feedback loops, CRM-style follow-up, and product iteration — underpin patient and user engagement and adherence in digital-health products.

What I Did
  • Owned product positioning, customer journey, CRM, service workflow, online/offline operations, and performance tracking.
  • Launched and iterated a WeChat mini-program for ordering, booking, and payment.
  • Analysed behaviour and feedback from 1,500+ WeChat community members and 1,000+ organic RedNote followers.
Actions
  • Used customer feedback and A/B testing to improve the user journey and increase online conversion by 10%.
  • Ran CRM-style follow-up and feedback loops to support retention and repeat visits.
  • Translated community feedback into product iteration, content strategy, and campaign decisions.
Relevant Skills
0-to-1 productCustomer journeyCRMA/B testingCommunity operationsProduct iteration

Case 05

Operational KPI Dashboards & Reporting

SQL / Excel / KPI Dashboards / Business Analysis / Management Reporting
Context

At Only Education, multiple campuses needed clearer visibility into enrolment, renewals, refunds, attendance, revenue, and course consumption for weekly and monthly management reporting.

Business / Operational Value

Turned fragmented campus data into structured reporting and recommendations — the data-to-decision foundation that the healthcare dashboards build on.

What I Did
  • Built SQL/Excel dashboards and KPI trackers covering enrolment, renewals, refunds, attendance, revenue, course consumption, and campus performance.
  • Analysed student behaviour, retention, conversion, and operational trends across campuses.
  • Prepared PowerPoint reporting for sales and management teams.
Actions
  • Tracked weekly and monthly KPIs across multiple campuses.
  • Translated fragmented operational data into management reporting, business insights, and follow-up recommendations.
Relevant Skills
SQLExcelKPI dashboardsBusiness analysisManagement reportingData storytelling

Case 06

Hands-on Technical Lab (Edge Delivery)

Technical Proof / DNS / HTTPS / Edge Delivery / Reproducible Walkthrough
Context

A small supplementary lab, kept to show the hands-on technical ability behind the data products: I built a custom-domain edge-delivery demo to test DNS, HTTPS/TLS, redirects, cache behaviour, and response headers.

Business / Operational Value

Supplementary technical proof that I can build, test, and explain web-delivery and infrastructure concepts — the practical skills behind deploying dashboards and data tools.

What I Did
  • Configured a custom-domain edge delivery demo.
  • Tested DNS, HTTPS/TLS, redirects, cache behaviour, and response headers.
  • Prepared a reproducible walkthrough for non-specialist readers.
Actions
  • Validated DNS routing and HTTPS responses with command-line checks.
  • Checked redirect status codes, security headers, cache-control, and edge cache behaviour.
  • Summarised the setup in a walkthrough written for non-specialist readers.
Relevant Skills
DNSHTTPSEdge deliveryHTTP headersTechnical walkthrough
Positioning
These cases form a focused portfolio for medical AI, digital health, healthcare analytics, and clinical-review-support roles. All medical AI work here is non-clinical proof of concept: it is not a validated clinical risk score, does not replace clinical judgement, and shows no patient-level or restricted data.