Work
Experiments
About
Resume
Designing An AI- Powered Dashboard for Malawi
Led by gHealth Research and University College Dublin, to address child pneumonia in Malawi. The project builds on a decade of work uniting PhDs, researchers, field experts, and health professionals to strengthen Malawi's health systems.
Our challenge was to design a user-centred AI dashboard that provides decision-makers with timely, actionable data where every decision can impact a child's life.
Role
⭐️ Product Designer
Design Systems
Research
Illustration
Documentation
Experience
Interaction
Team
Timeline
3 months
What we AIM to solve?
In Malawi, pneumonia remains the primary cause of childhood mortality. Health infrastructure suffers from fragmented data, slow reporting, and a lack of clinical decision support. District officers must interpret complex data under pressure, and current systems often overwhelm rather than enable, leading to missed early interventions. We aimed to close this gap by designing a dashboard that surfaces critical pneumonia trends, simplifies analytics with AI, and delivers clear, context-aware insights for urgent decisions.
By transforming scattered reports into cohesive, visual information, we aimed to help officials act confidently and allocate resources and strategies efficiently.
Setting up the Context
A strong foundation already existed, shaped by the research and lived experiences of previous contributors. The team spans continents from Ireland to Africa, Southeast Asia, and beyond reflecting a global partnership focused on collective impact. The innovation is fundamentally human. Malawi, one of the world's least economically developed countries, is taking a bold step by using AI to tackle pressing health issues.
In our very first meeting, a single measure of success was shared by partners that
“more children being able to attend school, free from illness. This metric is simple and human-framed every decision and design direction that followed."
Navigating Ethical and Legal Boundaries
Given the sensitive nature of this work, every element of our solution is guided by GDPR guidelines and bound by strict NDAs. The outcome showcased here uses a rapid prototype, carefully crafted to give an authentic sense of the experience while respecting data privacy and compliance.
Understanding the Challenge
We began with a clear challenge: Malawi’s health system was hindered by fragmented data and slow reporting, which impeded effective decision-making. To address this, we conducted detailed stakeholder interviews, half of which took place in Malawi and the other half globally, spanning district health officials, AI policymakers, and public health experts. Their perspectives confirmed the need for a dashboard that bridged urgent local realities with ethical AI standards.
Key Research Insights
Analysis of nearly a decade of epidemiological data revealed that under 1% of Malawian children with pneumonia received a complete clinical assessment, and only 16% had their respiratory rate measured. Despite technical advances, field users were overwhelmed: officials repeatedly called for clear geographic breakdowns, real-time outbreak alerts, and simple trend summaries to support fast, local responses. As one technologist put it: “We need to know who, where, and their epidemiological characteristics so that response can happen fast at the district level.”
Research Approach and Thematic Focus
We employed Reflexive Thematic Analysis on the interview data, tracking over 100 variables to identify five themes: district context, visual preferences, digital infrastructure, data freshness, and the quality of AI outputs. This living codebook kept our design grounded in user needs. Through iterative ideation, we focused on four main areas: trust, clarity, customisation, and feedback. The result was a modular dashboard comprising eight core modules, including a case explorer for granular analysis and an AI governance section for tracking accuracy and compliance.
From Ideas to Usable Solutions
We designed actionable user flows that enabled health officers to log in, scan for outbreaks, flag anomalies, and generate summary reports for ministry use. Instead of overloading the dashboard, we used a Value-Feasibility Matrix to prioritise core features, focusing on those that reduce uncertainty, enhance clarity, and suit Malawi’s low-bandwidth, offline-prone environment. These real-world constraints influenced features such as Excel-compatible exports and offline access.
Key pivots included transitioning from a mobile-first to a web-first architecture, driven by the workflow realities of ministry offices. Early prototypes tested trends visualisation, flagging, and data aggregation, providing usability insights that shaped each new version.
“The report also needs to have the GIS function... we can populate the Use the heat map to show the To show the hotspots based on the district and even the location.”
Public health and Information and Communications Technology technologist, Male, Malawi
“I think you see the Epi curve... I think that it's useful to look at the number of cases per month, then the heat map. I think it is a nice idea where you could see how they are distributed throughout the region, and then cases by district.”
Project Partner & Clinical Director of the gHealth Research Group, Male, Ireland
“Weekly IDSR report, data quality, completeness, timeliness.”
Public health and Information and Communications Technology technologist, Male, Malawi
“providing some alerts, so something like, not their message, but just like, you know, like alerting something in a way, like just as a warning.”
Research assistant in IT, Female, Ireland
“filters on the side for seasons and sites and age groups and things like that... your Epi curve is the same idea... It's tracking prevalence, incidence over time, and different years.”
Project Partner & Clinical Director of the gHealth Research Group, Male, Ireland
“using the GIS will be most powerful. If we can interactively click on a certain history, and then we know all the information. Or if the trend that we want to analyze, we can also base on those to know how we can relocate the resource.”
Public health and Information and Communications Technology technologist, Male, Malawi
“internet issues and the power issues are very common in Malawi”
Public health specialist, Male, Malawi
“mortality rates, admission rates, fatality rates" as key outcome indicators.”
Microbiologist and Immunologist, Male, Malawi
“Do we have enough medicine? In the facility or in the district, if it's not enough, then how do we relocate that?”
Public health and Information and Communications Technology technologist, Male, Malawi
“The safety, the system, that's safe, how much risk it's kind of a deal risk... the quality of the data, all those things will kind of be another kind of dimension.”
Information Systems sector, Female, Ireland
“Having information on vaccinations would also be useful... having the dashboard have the usage of antibiotic, and biomarker trends.”
Research assistant in IT, Female, Ireland
“They want to see if this changes over time, whether it is accurate overall, and whether predictions vary by age, gender, site, or season. Filters can be used to identify sites that are not performing well with the AI score.”
Project Partner & Clinical Director of the gHealth Research Group, Male, Ireland
“If it is put in pictorial form, maybe like bar charts, it would be very easy for me to check the trend.”
Public health specialist, Male, Malawi
“There are some kinds of metrics that measure fairness... the sensitivity between the best and the worst served demographic group.”
Information Systems sector, Female, Ireland
“It also highlights bias and fairness, ensuring the tool isn't biased against certain seasons or districts. These are part of the AI governance framework.”
Project Partner & Clinical Director of the gHealth Research Group, Male, Ireland
“For example i am based in Mzuzu near Tanzania and is very cold while a different place could be hot right now which could affect the cases”
Public health specialist, Male, Malawi
Clinic Level
+
+
Diagnostic Turnaround - CUSUM
True/False Positive
Antibiotic Usage & Resistance
District Level
Heat‑map View
Antimicrobial Resistance Mapping
SARI Tracker
Ministry Level
National Epidemic Curve
Case Demographics Breakdown
AI‑generated Summary Reports & Governance Logs
Oxygen & Bed Occupancy
Referral Tracking
Supply Chain Monitoring
Alerts (Mail/WhatsApp)
Weekly Reports / IDSR Metrics
Decision-Making and Trade-Offs
Design decisions often strike a balance between complexity and simplicity. Should we enable deep filtering or keep the dashboard straightforward? We grounded every choice in speed, clarity, and flexibility, so health officials could drill down for detail only when needed. As one clinical director said, “We don’t want to bump up the user with too much information, but the system should let us drill down for detail when we need it.” All metrics, accuracy, delays, and user actions were tracked to validate usability and guide future improvements.
Delivering the Solution
We delivered a rapid prototype through iterative design sprints and user validation, focused on integration, usability, and transparency. Built as a lightweight overlay for Malawi's existing ICHIS and DHIS2 systems, the dashboard lets health workers and officials maintain familiar workflows while gaining access to AI-powered analytics. Data pipelines processed 100+ clinical variables but distilled insights by district, pathogen, and demographic group. Predictive maps and heatmaps simplified resource allocation decisions. Recognising the limitations of power and the internet, we developed offline-first functionality that syncs automatically, ensuring that no records were ever lost.
Health workers continue to use DHIS2, REDCap, E‑Vax, and LMIS exactly as before. Forms, workflows, and user experiences are unchanged. We accept data from multiple sources without modifying front-end processes. Data entries proceed normally in all existing apps. Offline saves still rely on the app's built-in mechanisms.
Data Collection
All the data is automatically funnelled into the cloud, where it’s batched and cleaned on a fixed schedule, then sent straight into our Biotope ML model for risk scoring and into DHIS2 for broader analytics - no manual steps required and no changes to how data is entered in the field.
Data Processing
Our stakeholder engagement revealed that mobile access was not essential for the current operational model. Most dashboard consultations occur during planned data reviews or official coordination meetings, where desktop infrastructure is already available. This allowed us to deprioritize mobile optimization in the initial release phase.
Data Visualisation
Asynchronous SNS/SQS
FHIR-compliant JSON
Scheduled ETL Layer
Data Collection
Capture App
Health Management Information Service
(Patient Data)
REDCAP
(Patient Data for Testing and Training the AI model)
Primary Input
Secondary Inputs
Logistics Management Information Service (LMIS)
(Drug and Antibiotic Stock)
E-Vax
(Vaccination Data)
FHIR encoded as protobuf file
S3 + CloudFront
Background Sync
Background Sync
Medical Data Delivery
JSON file
System
Data Logging
Caching Layer
Local Storage
Dashboard
BIOTOPE
Data Processing
Data Presentation
ML Model for Prediction
AI Model
DHIS2 Analytics
Data Analytics
Data Flow
Data Sync
Compliant Features
Impact on Decision Making and Practice
The dashboard turns complex data into actionable, context-driven choices. Officials can now visualise pneumonia clusters, monitor antibiotic use, and compare outbreaks quickly, supporting faster interventions. Tools like the case explorer and filterable heatmaps enable leaders to target high-risk districts, allocate resources effectively, and track the impact over time. In pilots, staff generated exportable reports in half the time it took previously; dashboard alerts directly triggered resource reallocation. Feedback features enable users to flag and comment in-tool, continually improving clarity and surfacing needs.
In Malawi, every data point represents a life.
Our dashboard aims to support early interventions and resource allocation to protect children like those pictured here.
This stage made me realise how fragile a good design can be when ideas move from paper to structure. Decisions that felt small, like where to place a filter or how to layer data, suddenly carried weight because they changed how people would interact with the system. I noticed that some user needs clashed with technical constraints, and those moments became our best opportunities to rethink priorities rather than compromise quietly. It reminded me that architecture is not invisible work; it sets the tone for trust and usability. These lessons shaped how I approached the next phase, where the challenge was to bring fidelity to those ideas without losing the clarity we had worked so hard to build.
Reflections and Forward Insights
The project highlighted both the capabilities and the limitations of digital health tools. Lessons from early prototype failures showed that trust and context are vital for adoption. Prioritising web-first architecture and contextual visual cues was driven by honest user feedback. Impact came not just from advanced models, but from deep listening and adaptability. The collaborative spirit, involving researchers from four continents and communities across Malawi, was central to maintaining the work's honesty and urgency.
Copyright © 2025 Kamaljeet Singh. All rights reserved.
Curious to know more?
Design with Kamal
Connect with me at kajeesingh@gmail.com or lets dive into some conversation over chai
Download CV
Curious minds never stop!
Unlock the full story — get the complete document.
© 2025 UCD. Confidential and protected under NDA. No unauthorized use or sharing.
Download Document