Building LDiHK: Our Journey at the EuroTech x Hong Kong Talent Exchange Hackathon 2026
Earlier this year, I had the opportunity to compete in the EuroTech x Hong Kong Talent Exchange Hackathon 2026. Joining forces with three incredible teammates, we tackled a pressing issue in modern healthcare: the intersection of digital behavior and mental health.
Our solution, Life Data Intelligence Hong Kong (LDiHK), was born from a simple but powerful premise: we can catch the warning signs of mental health crises earlier by analyzing the data people already generate every day.
Here is a look behind the scenes at what we built, the tech that powered it, and my role in bringing the MVP to life.
The Problem: A System Overwhelmed
Hong Kong is facing a silent mental health crisis.
Recent studies show that nearly 46% of people experience symptoms of mental disorders, and a stable psychiatric referral can take up to 44 weeks.
By the time patients get the help they need, they are often already in crisis. Simultaneously, decades of peer-reviewed research have drawn a clear link between social media behavior and depression risk. Heavy daily use, fragmented attention spans across multiple platforms, and late-night scrolling that disrupts circadian rhythms all measurably increase the odds of internalizing problems.
We wanted to bridge the gap between these two realities.
The Solution: Life Data Intelligence Hong Kong
Our pitch was to turn digital tracking data into proactive care. We developed a platform that allows users to voluntarily upload their exported social media usage data (via GDPR takeout requests from platforms like YouTube, Instagram, and LinkedIn).
The platform normalizes the data and scores the user's risk for depression using a mathematical model. But LDiHK isn't just a diagnostic tool; it's a B2B service tailored for health insurance providers.
The Business Model
- Insurers License the Model: We sell our machine-learning classification model to insurance companies.
- Voluntary Data Sharing: Insurers offer premium discounts to members who voluntarily share their social media usage data.
- Proactive Intervention: Insurers use our risk assessment to recommend early prophylactic treatments.
- The Win-Win: Patients receive early care, and insurers cut the massive long-term costs associated with untreated mental illness.
My Role: Frontend, Infrastructure, and Pitching
Working in a team of four meant we all had to wear multiple hats. I primarily owned the frontend architecture and shared responsibilities for our hosting and tech stack, while also contributing to our core ideation and the final pitch.
1. Architecting the Frontend and Tech Stack
Our architecture needed to handle data ingestion, normalization, enrichment, and querying—all while providing a visually intuitive dashboard for the user. I built out the frontend to ensure a seamless flow from the moment a user uploaded their ZIP file to the presentation of their personal insights and population benchmarks.
Our hosted architecture relied on a modern, scalable stack:
| Stack Layer | Technology | Primary Function |
|---|---|---|
| Frontend | Netlify | Handled the UI and user-facing analytics. |
| Backend | Render API | Processed uploads and ran enrichment workers. |
| Database | Supabase Postgres | Stored pre-processed data and reference demographics. |
| Storage | AWS S3 | Handled the temporary storage of large ZIP file uploads. |
A key technical challenge was dealing with raw, fragmented data from different platforms. I helped design a pipeline that could take in various log files, normalize them into a unified standard, and display them intuitively. For instance, we utilized the YouTube Data API via background workers to enrich watch histories with precise video durations, giving our model accurate screen-time metrics to work with.
2. The Mathematical Groundwork
To ensure we were building something grounded in reality, our risk scoring logic was built entirely upon independent, peer-reviewed studies (such as data from the UK Millennium Cohort Study and JAMA Psychiatry).
We engineered a multivariable logistic regression framework to fit a continuous predictive function through our data points. Because digital stressors affect a 16-year-old very differently than a 60-year-old, we made the backend decision to split our logic into three separate models: adolescents, young adults, and older demographics. While this increased backend complexity, it was non-negotiable for delivering clinical accuracy.
3. Shaping the Pitch
Beyond the code, I helped refine our narrative for the judges. We needed to prove that LDiHK was not just technically sound, but legally compliant, socially responsible, and financially viable.
- Ethical Data: We heavily emphasized that our platform does not scrape data—it relies entirely on opt-in, consensual data sharing.
- Strategic Launchpad: We also highlighted why Hong Kong was the perfect starting point. Sitting at the intersection of Western (WhatsApp, Facebook) and Chinese (WeChat, Xiao Hong Shu) digital ecosystems, it offers a unique environment to train robust, cross-cultural models before expanding across Asia.
Looking Back
Building LDiHK under hackathon time constraints was an intense lesson in prioritization. We had to balance complex data engineering with creating a polished, user-friendly MVP. Seeing our data pipeline successfully ingest raw social media logs, apply our regression framework, and output meaningful, actionable health insights was an incredibly rewarding moment.
The hackathon reinforced for me that the most impactful tech solutions don't just solve engineering problems—they align human incentives. By creating a system where users get cheaper insurance and insurers save on long-term medical costs, we built a product that actually has a path to real-world adoption.