Hey HN, we want to share HelixDB helix db, a project my team is building to solve the fragmentation in AI data infrastructure. Why a hybrid? Modern AI agents require both semantic similarity and relational context, yet developers are currently forced to cobble together separate vector and graph databases. This leads to latency, synchronization hell, and increased infrastructure costs. HelixDB solves this by natively combining property graph engines with approximate vector search into a single operational layer. Here are a few ways this improves development: - Knowledge Graph RAG: You can perform multi-hop reasoning by traversing entity relationships while simultaneously filtering by semantic similarity, allowing your LLM to answer questions that span complex causal chains. - Fraud Detection: Use vector embeddings to identify suspicious patterns while traversing user transaction graphs to find connected illicit actors in one query. - Recommendation Engines: Combine user interest embeddings with social graph traversals to serve highly personalized content that respects explicit user relationships. Many developers are understandably skeptical of 'yet another database' claiming to handle everything. We went ahead and built it in Rust anyway, specifically to avoid the memory overhead and concurrency bottlenecks found in JVM-based alternatives. By utilizing a custom LSM-based storage engine, we achieve performance that is, according to our initial benchmarking, on par with Pinecone for vector throughput while performing relationship traversals up to two orders of magnitude faster than standard deployments of Neo4j. If you want to see how this works in practice, check out our getting started guide or watch our technical walkthrough video. If you’d like to try it out in a simple RAG demo, you can follow this integration tutorial. Many thanks! Comments and feedback welcome!