Work → Case Studies → Wellness Buddy Thailand — LLM-Powered Wellness Recommender
Wellness Buddy Thailand — LLM-Powered Wellness Recommender
I built Wellness Buddy because the wellness economy in Thailand is already a multi-billion-dollar opportunity, yet many travelers still make expensive decisions with low confidence and incomplete context. I wanted to turn fragmented reviews and marketing language into something people could actually trust when they are about to spend real money on their health.
Behavior Lens
Great products do not just organize information. They reduce anxiety at the moment of decision.
Overview
I built Wellness Buddy because the wellness economy in Thailand is already a multi-billion-dollar opportunity, yet many travelers still make expensive decisions with low confidence and incomplete context. I wanted to turn fragmented reviews and marketing language into something people could actually trust when they are about to spend real money on their health.
Role: AI Product Developer & Researcher
Timeline: 2024–Present
Tools: Python · NLP · LLMs · Next.js
The Problem
Big Market. Confused Buyer.
Thailand is one of the world's strongest wellness destinations, but abundance created a paradox: more choices did not create more confidence, they created more hesitation. In practice, most travelers were making high-stakes decisions with low-quality guidance.
I kept seeing the same behavior pattern in user language and browsing paths. People were not choosing the most suitable program; they were choosing the option that looked safest, sounded familiar, and required the least mental effort to justify to themselves.
Under the surface, 430+ wellness programs were described inconsistently across websites, while 6,400+ reviews remained unstructured and underused. The information existed, but the decision architecture was broken.
Research
Listen to Behavior, Not Hype
- I analyzed 6,400+ traveler reviews from Google Maps and TripAdvisor across 165 wellness centers to capture real preference language rather than brand copy
- I mapped language patterns into 6 wellness dimensions: physical, mental, spiritual, emotional, social, and environmental
- I identified a repeated mismatch: travelers describe goals in personal, emotional terms, while providers market programs in operational categories
- I validated these findings academically and published them in IEEE ICSEC 2024
Solution & Build
From Research to Real Product
- I built an NLP pipeline to classify review text and extract sentiment signals by wellness dimension
- I trained an intent-matching model that reached 90% classification accuracy for mapping user intent to program categories
- I added an LLM layer to convert technical matching logic into plain-language recommendations people could understand quickly
- I published the extended framework in IEICE Transactions 2025
- I shipped wellnessbuddythailand.com as a live production product that connects research rigor with real-world traveler decisions
Product Impact
The Impact Funnel We Built
As a Product Manager, I wanted the case to show more than technical accuracy. I mapped impact as a decision funnel: signal quality, recommendation quality, and user confidence at the point of choice.
At the top of the funnel, we transformed 6,400+ unstructured reviews from 165 wellness centers into usable behavioral signals. In the middle, the model reached 90% intent-classification accuracy, which improved recommendation relevance. At the bottom, recommendations were translated into plain language so travelers could decide faster and with less anxiety.
That structure turned academic output into product output. Not just a model that works in a paper, but a recommendation flow people can trust when money, health goals, and travel expectations are all on the line.
- Input quality: 6,400+ reviews converted from noisy text into structured traveler intent
- Matching quality: 90% classification accuracy across wellness intent categories
- Decision quality: recommendations presented in human language, not technical labels
- Business relevance: live product deployed in a multi-billion-dollar Thailand wellness market
Outcome
Live product serving international wellness travelers in Thailand
2 peer-reviewed publications tied directly to shipped product decisions
End-to-end research-to-product execution completed as MSc thesis
Closed a real decision gap in a multi-billion-dollar tourism category
Learnings
I learned that people rarely choose the mathematically best recommendation if the explanation feels abstract or cold. They choose the option that helps them feel seen, reduces uncertainty, and makes a difficult decision feel clear in under ten seconds. Accuracy builds credibility, but clarity is what moves behavior.