Building a Real-Time Recommendation Engine for a Video Game Store with VelesDB
The article describes how to build a real-time recommendation engine for a video game store using VelesDB, a local-first vector, graph, and columnar database written in Rust.
Why it matters
This approach to building a real-time recommendation engine can be useful for video game stores and other e-commerce applications that want to provide personalized recommendations without the latency and cost of external services.
Key Points
- 1VelesDB is used to store product embeddings and user behavior data
- 2The recommendation engine handles anonymous prospects and authenticated customers differently
- 3For anonymous users, content-based filtering is used to recommend similar products
- 4For authenticated customers, the engine combines graph-based and vector-based recommendations
Details
The article explains the problem with classic recommendation systems that either rely on external services or pre-compute stale recommendations. The proposed solution uses VelesDB, a lightweight database that runs locally next to the application. For anonymous users, the engine uses content-based filtering by storing product embeddings in the 'products' collection and performing vector search. For authenticated customers, the engine combines graph-based recommendations (using the 'user_behavior' graph) and vector-based recommendations (from the 'products' collection). This allows the engine to react to the current user session in real-time without external dependencies.
No comments yet
Be the first to comment