Format
Personal web prototype built with Python and browser automation.
Case study
Building a Supplement Recommendation Prototype from Product and Review Data
A personal project combined user profile inputs with product and review data to explore how a recommendation flow could be made more concrete.
Published Mar 23, 2026 Updated Mar 24, 2026
Overview
The project started from a simple question: how should recommendation logic stay grounded?
Context
The idea was to build a web application that suggested dietary supplements based on profile inputs such as age, gender, lifestyle, habits, goals, and pregnancy status.
I wanted the recommendation flow to be informed by actual product and review data rather than by a fixed list of generic suggestions.
Approach
I collected public product information and review data, then used that material to shape a recommendation prototype that could connect user profile inputs to a more concrete product set.
The project was exploratory rather than clinical. The point was to build a usable recommendation flow, not to present medical advice.
Workflow
The prototype linked questionnaire inputs to a searchable product base.
Profile inputs
The interface captured user traits and goals that could influence which supplement categories looked relevant.
Public product data
Public product information and review text were collected so the recommendation flow could be tied to a real catalog instead of abstract examples.
Recommendation logic
The prototype connected the profile and catalog layers so the user could move from questionnaire answers to a narrower set of options.
Results
The useful outcome was an end-to-end research prototype.
Type
PrototypeExplored a recommendation flow rather than a finished consumer product.
Inputs
Profile + catalogCombined user profile fields with public product information.
Stack
PythonUsed Python and browser automation for data collection and prototyping.
Takeaways
Recommendation projects improve once the data and the user flow are designed together.
What mattered
The interesting part was linking messy real-world catalog information to a simple questionnaire-driven interface.
Why it was useful
The project provided a concrete way to think about how product data, user traits, and recommendation logic should connect in one flow.
Related
Continue with connected pages
Case studies and methods that connect to the same operational questions.
Case study
Sales Forecasting Competition
Won a public forecasting competition by matching the preprocessing and forecast logic to the competition's asymmetric scoring rule.
Case study
Tweet Sentiment Tracker
Built a public-data analysis project that collected historical tweets, stored them in PostgreSQL, and scored sentiment over time.