Even clinical teams struggle to stay up to date. For patients, it is nearly impossible.

People living with serious or chronic health conditions want the most up-to-date information so they can make the best decisions about their care. However, for most conditions there are over 1,000+ new developments every week across research papers, clinical trials, expert talks, webinars, podcasts, and patient communities.

The challenge was not simply access to information. It was how to continuously surface the most relevant new insights for each individual as their health journey evolves.

We tested the idea with one generic email — before building anything.

We tested this idea early by creating a simple email called “What’s New This Week in Your [insert disease name].” The email was generic and not personalised, yet it consistently achieved very strong engagement:

This led us to design Turny, an adaptable personalised weekly health insight system.

I designed the system 0→1.

I worked closely with my lead engineer to shape system behaviour and architecture, including how health profiles interact with the knowledge dataset and how insights are ranked and surfaced. Throughout development I helped identify edge cases across the insight pipeline and refine the product logic from data ingestion through to personalised insight delivery.

Designing an insight updating system

We onboarded users to see impact quickly

• Only 5 quick-fire onboarding questions
• Simplified onboarding using voice-to-text
• 1st update generated before sign-up to trial
• Deep search of articles relevant to the user’s specifics
• Early indication that the user’s priorities matter!

And provided a weekly cadence of continued insights

• 10–15 items of high-value relevant information
• Sent same day, same time each week
• Encouraged to provide feedback & update profile
• Multi-modal digestion of the update overview

Product outcomes and key learnings

There have been several consistent patterns in how patients engage with their health updates:

People researching complex health questions rarely rely on a single source.

Important insights are scattered across many places, including:

Today these sources exist in separate ecosystems. Most information systems prioritise academic sources while ignoring expert discussions and patient experiences, and do not optimise for recency.

As a result, people are forced to piece together fragmented information across many platforms, while remaining vulnerable to misinformation or oversimplified advice.

All sources matter. And recency matters.

1. All sources matter: valuable insights can emerge from research, expert discussions, and patient communities. What patients are experiencing in real life often takes years to appear in formal research, yet these experiences can highlight emerging treatment patterns or side effects much earlier — alongside the importance of reliable evidence next to these patient discussions.

2. Recency matters: patients care deeply about what is happening right now. What a world leader said about a new treatment on a podcast yesterday matters.

I led the design and development from concept to early deployment.

The project required bridging AI capabilities, medical information systems, and user needs in complex health decision-making.

Designing a multi-perspective AI deep-dive system

When a user asks a health question, the system does not immediately generate an answer. It first captures key context signals, then generates a structured deep-dive report by retrieving evidence from a structured health knowledge index — using semantic retrieval and ranking models that prioritise relevance and recency.

A different approach to Q&A

Health questions rarely have a single definitive answer. Research may be incomplete, expert opinions evolve, and patient experiences often surface emerging patterns years before they appear in formal studies.

Instead of compressing everything into one summary, the system was designed to surface multiple perspectives and allow users to explore the evidence themselves. In some cases the system explicitly acknowledges uncertainty.

Product challenges overcome

“Hands down the best tool I have ever used that deals with complexity of health answers — no one vanilla answer, but actually tells you there is no answer, or limitations of what is.”

— health clinician

Product outcomes and key learnings

Thousands of structured health pages published, creating an organic acquisition channel for the whole platform. Direct source navigation — jumping to the exact moment in a podcast — measurably increased trust and exploration.

For most conditions, no comprehensive list of treatments exists.

In cerebral palsy, families are typically introduced to a small number of therapies through their clinical care team. Yet across research literature, specialist clinics, and emerging therapies, hundreds of treatments exist — scattered across academic papers, conference talks, and clinical practice, with little clarity about how widely a treatment has been studied or how strong the evidence may be.

The challenge was not simply accessing research papers. The deeper problem was the absence of a structured treatment taxonomy — medicine rarely maintains a comprehensive, up-to-date map of treatments for a condition. At the same time, families evaluating treatments value multiple perspectives: research evidence, expert interpretation, and patient experiences each provide different signals.

I founded and led the development of this project.

After attending a leading cerebral palsy research conference and identifying the gap in how treatment knowledge was organised, I designed the concept for a system that could map the full treatment landscape for the condition. I then assembled the project team and secured research funding to build the system.

System design: mapping the treatment landscape

The system combined research evidence, expert insight, and patient experience: it analysed 32M research abstracts, identified 250+ treatments, and organised them into a structured taxonomy so families could explore treatments by both type and treatment goal.

Machine learning models analysed 32M PubMed abstracts to identify cerebral palsy treatment studies and extract key signals such as study type, participant size, outcomes, and effectiveness indicators. To support model training and validation, three practising physicians were engaged part-time to review abstracts and help define classification criteria aligned with the GRADE evidence framework.

This allowed the system to translate complex research literature into clear evidence-strength indicators for users.

Product experience: exploring the treatment landscape

The platform mapped more than 250 cerebral palsy treatments into a structured taxonomy, so users could explore therapies by both treatment type and treatment goal — discovering treatments beyond those typically introduced in clinical care.

Each treatment page combined research evidence indicators, expert interpretation, and patient treatment experiences to support more informed decision making.

Product outcomes and key learnings

People living with chronic conditions run out of ideas.

Three things we knew going in

I led the product from concept to launch.

I worked closely with engineering on product behaviour and feature design, translating user needs into practical mobile experiences.

System design for daily content delivery

An app that tells you to stop

Product outcomes and key learnings

This product showed strong product-market-fit signals — and clarified the operational limits of a high-touch daily model: