Atlas
A private place recommendation engine that aggregates Google Takeout history and group taste profiles into explainable suggestions.
What I Built
Atlas is a web application that imports personal Google Takeout place files, shares them inside a small invite-only circle, and generates recommendations grounded in your group’s taste rather than public reviews.
Why It Matters
Public local discovery engines optimize for popularity and star averages, which often leads to generic recommendations. Atlas tests the idea that matching taste profiles and saved locations within a trusted group of friends produces more relevant results.
System Design
The application runs on a Node.js and Express backend with an SQLite database, serving a React frontend. To generate recommendations, the system constructs a prompt containing:
- Taste Profile: Mapped categories and favorite neighborhoods derived from takeout imports.
- Social Proof: Saved places and matching entries from friends in the circle.
- Places Data: Live metadata fetched and cached from the Google Places API.
An LLM synthesized explanation (using DeepSeek via OpenRouter) compiles these inputs into a recommendation that explains why a spot is suggested. A validation layer sanitizes queries to block prompt injection and prevent hallucinating fake location features.
Screenshots & Visuals
