Commit

Suprnova 4.3.0 has been released I've been building something pretty ambitious into Suprnova, my Laravel-inspired Rust framework. Three tightly integrated engines: Search, Recommendations, and a Knowledge Graph. And you don't need a collection of external services to use them. 🔎 1. Search Engine Suprnova includes a full-featured search engine built directly into the framework. Full-text search with relevance ranking Typo tolerance, synonyms, and highlighting Filtering, faceting, sorting, and geospatial search Federated search across multiple indexes Hybrid lexical + semantic vector search Optional local AI embeddings, including multimodal support Pluggable backends, including your existing SQL database and Qdrant Declare your Eloquent models searchable and Suprnova handles the indexing, retrieval, ranking, and model hydration. No Elasticsearch, Meilisearch, or hosted search service required. 🎯 2. Starter Recommendation Engine Not every application needs a complex machine-learning recommendation system. Suprnova includes a recommendation engine that works without embeddings or an external AI service. It learns from interactions such as views, clicks, likes, shares, purchases, skips, and hides. It combines several recommendation strategies: Collaborative filtering based on shared engagement Attribute affinity using searchable document facets Popularity based on actual activity Freshness and recency Optional semantic similarity The engine progressively falls back through these strategies when stronger recommendation signals aren't available. Recommendations are explainable, respect access permissions, and are backed by an immutable interaction history. You can build personalized feeds and content discovery without deploying an ML infrastructure. 🕸️ 3. Knowledge Graph Recommendation Engine Here's where things get particularly interesting. Suprnova also includes a knowledge graph engine that connects searchable documents through shared engagement and explicitly declared relationships. Instead of merely asking, "What did similar users like?", it can evaluate relationships between content, people, and their interactions. The graph combines: Multi-hop relationship traversal Shared engagement patterns Semantic similarity from existing search vectors Social proximity Freshness and negative feedback Trust signals and diversity controls It ranks recommendations dynamically from the graph, with verifiable provenance linking recommendations to the evidence behind them. Access controls apply at every hop, not just to the final result. And if the graph doesn't have sufficient evidence? It automatically falls back to the starter recommendation engine. The graph can run directly in your existing SQL database or use Memgraph. Here's what makes the architecture interesting: These aren't three separate products glued together. Search provides the indexed documents and optional vectors. The recommendation engine captures interactions and builds an explainable behavioral history. The knowledge graph turns those documents, interactions, and relationships into richer contextual recommendations. The engines share data, scopes, and infrastructure. Search impressions and validated clicks can even feed directly into recommendation events. Start with ordinary full-text search. Add personalized recommendations. Enable graph-based discovery as engagement grows. All inside the same Rust framework, with a consistent API and no mandatory third-party SaaS dependency. That's the kind of infrastructure I think modern application frameworks should provide out of the box. Built with Rust. Powered by Tokio. #RustLang #OpenSource #Suprnova https://suprnova.app/docs/recommendations… https://suprnova.app/docs/search