PlayBeacon PoC: AI Game Recommendation Engine

Proof of concept that validated the core recommendation engine powering PlayBeacon.

Before committing to a full product architecture, this prototype tested whether AI-powered personalized game discovery was feasible at scale using vector similarity search. It was, and it became the technical foundation for PlayBeacon.

---

What This Validates

• pgvector similarity search produces relevant, fast game recommendations
• OpenAI embeddings effectively capture game "feel" from titles and descriptions
• A Steam-style discovery queue UX translates naturally to the Roblox context
• The full pipeline (crawl, embed, store, recommend) can run end-to-end

---

Architecture

playbeacon-poc/
├── backend/
│   ├── crawler/     # Indexes Roblox games by keyword
│   ├── embeddings/  # OpenAI embedding pipeline to pgvector storage
│   └── api/         # FastAPI recommendation endpoints
└── frontend/        # React swipe interface (discovery queue)

---

Tech Stack

---

Key Components

• Game Crawler: automatically indexes Roblox games by keyword
• Embedding Pipeline: OpenAI embeddings stored in PostgreSQL via pgvector
• Recommendation API: FastAPI + cosine similarity search for personalized results
• Discovery UI: React swipe interface modeling Steam's discovery queue

---

What Came Next

This PoC graduated into PlayBeacon, a full production app for teens and Roblox players, live on iOS and Android.

---

Built by Watchlight Interactive