Roundtable summary and recording: Introducing Box Search 3.0

Thanks to Shuang Li and Veerender Nallamala for this weeks overview of Box Search 3.0 and also thanks to our customer guests who joined and shared in the Q&A. You can find the summary below, and full recording at the bottom.

Summary

Box Search 3.0: A New Foundation for Performance

Veerender explained that Box Search 3.0 represents a major architectural shift in the platform's search infrastructure, designed to address common customer feedback regarding speed, the inability to find recently created files, and result relevance.

The new version, which has been rolled out to all customers (GA), focuses on three core improvements

1. Quick Search Speed: A live demo compared the old and new infrastructures on a large-scale system with a billion files. The time for search suggestions to appear in 'Quick Search' was reduced from approximately 8 seconds to just 3 seconds.

2. Full Search Reliability: The demo showed that complex 'Full Search' queries, which previously timed out without returning results on the old infrastructure, now successfully complete in a few seconds on the new system. This represents a 4-10x improvement in latency.

3. Real-Time Indexing: Veerender demonstrated the elimination of the previous 15-minute lag between a file's creation and its availability in search results. A newly created file was shown to be instantly discoverable, resolving a significant user pain point.

The Future Roadmap: Intelligent and Context-Aware Search

Shuang presented the forward-looking roadmap for Box Search, highlighting three key initiatives that build upon the new architecture:

1. Hybrid Search: This will combine traditional keyword matching with semantic search, allowing Box to understand the user's intent and the meaning behind a query. Users will be able to use natural language (e.g., "find the confidentiality agreement we signed last quarter") to find documents, even if the exact keywords are not used.

2. Multi-modality Search: Acknowledging that much of an organization's knowledge is in non-text formats, this feature will make images, videos, audio files, and scanned PDFs fully searchable. The system will be able to identify objects, text, and concepts within these files without the need for manual tagging.

3. Context Retrieval Engine: Aimed at developers and enterprise AI applications, this engine will provide a new API that returns precise, relevant text chunks and passages from documents instead of entire files. This will make it much more efficient for AI agents and custom applications to find and use specific information stored in Box.

Key Insights and Q&A

The session included a robust Q&A segment with valuable customer feedback and questions. Key discussion points included:

• Search Defaults: Several users expressed frustration that the search bar defaults to searching within the current folder rather than searching everywhere, requesting a change or an admin toggle for this behavior.

• Character Limits: The 10,000-character limit for full-text indexing was a significant concern for users with long documents. The product team confirmed it is on the roadmap to lift this limit and offered to evaluate high-priority cases individually.

• AI and Search: Customers asked about the ability to exclude certain folders from AI-aided search and how existing metadata will be integrated into future AI search capabilities. The team confirmed that all search, including AI, respects user permissions.

• Metadata Search: It was clarified that global search does look for matches within metadata field values, though the relevance ranking might not always place these results at the top.

Wrap up

The roundtable concluded with a poll revealing that 'Multi-modality Search' was the most anticipated future feature among attendees. 

Full Recording 

Full Slide deck

Slide deck is here