Databricks has developed a new architecture for its Knowledge Assistant to improve enterprise search capabilities. The system, initially called Instructed Retriever and later refined to Instructed-Retriever-1, addresses limitations in traditional retrieval-augmented generation (RAG) by enabling the system to follow complex instructions, such as exclusions and specific formatting, rather than just matching keywords. This redesign was further optimized for speed by implementing parallel search instead of sequential retries, allowing for faster and more accurate responses to intricate user queries. AI
IMPACT Enhances enterprise search by enabling complex instruction following and faster retrieval, improving user interaction with internal knowledge bases.
RANK_REASON The article details the architecture of a specific product, Databricks Knowledge Assistant, focusing on its technical implementation for enterprise search.
- Amazon Web Services
- Apache Spark
- Azure
- Databricks
- Delta Lake
- Google.Cloud
- Knowledge Assistant
- Python
- retrieval-augmented generation
- SQL
- Vector Search
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →