This question came up in our recent webinar Q&A [video below👇].
The short answer: Yes. Apache Doris can replace your existing vector store (ChromaDB, Pinecone, Milvus...), but your chunking, embedding pipeline, and application logic stay exactly as they are.
A lot of RAG systems infra today look like this:
- Postgres for structured data
- Pinecone/ChromaDB/Milvus/Weaviate for vectors
- Some even adding Elasticsearch for keyword search
- Your app stitches results together
But what if "clients want to query their database with an LLM, not just text, but structured and unstructured data together?"
When your vectors, keywords, and metadata live in different systems, it's difficult for you to do searches like this efficiently: "find Python engineers in San Francisco hired in 2024 with similar backgrounds to this resume."
But with Apache Doris, a real-time database that now support hybrid search and vector search, you can do those searches in one SQL query, in one database, using one unified system.
If you're running RAG in production, juggling multiple databases, and facing cost and performance issues, it might be worth asking: what if you didn't have to?
🔗 See how ByteDance uses Apache Doris' hybrid search to cut down vector search cost: https://www.velodb.io/blog/bytedance-solved-billion-scale-vector-search-problem-with-apache-doris-4-0?utm_source=linkedin
🔗 Watch the webinar in full: https://www.youtube.com/watch?v=kKiXWNWZYVc
Top comments (0)