ExaMeld’s Search module provides enterprise-scale file discovery through a combination of high-performance keyword indexing and AI-powered semantic search. The system enables rapid location of files across millions of entries using both exact keyword matching and conceptual search.
Index Management
Index definitions specify target locations (file system paths, UNC shares, or network mounted directories) and file patterns for inclusion. Individual indexes can be selectively enabled or disabled without reconfiguration, facilitating context switching between different monitoring scopes. Cached indexes store compressed data for rapid reuse across multiple searches.
Dual Search Engines
The Fast search engine provides efficient keyword and pattern matching, excelling at precise queries with known terminology. The Semantic search engine matches files based on conceptual similarity, enabling discovery when exact keywords are unknown — for example, searching for “authentication” can surface files about “login” and “credentials.” Both engines operate simultaneously, with results merged according to configured weighting.
Hybrid Scoring
Search results combine relevance scores from both the semantic and keyword engines. A configurable weight parameter blends these scores, enabling users to bias toward conceptual or exact-match relevance based on the nature of their search. Similarity thresholds filter out low-confidence results, keeping result sets clean and focused.
Search Interface
A browse mode enables exploration of indexed locations without active queries, displaying directory hierarchies and file listings. Animated indicators provide visual feedback during index operations. Index metadata display shows index size, file count, last update timestamp, and other statistics informing search performance expectations.
Results
Results are presented in a sortable grid with configurable columns including filename, path, size, and relevance score. Relevance visualization through color gradients or score bars enables rapid assessment of result confidence. File sizes are formatted for human readability (KB, MB, GB) rather than raw bytes.
Size Filtering
Minimum and maximum file size filters narrow result sets to files within specified ranges, supporting scenarios where file size correlates with relevance (e.g., searching for compiled binaries while excluding documentation). Filtered sub-indexes optimize storage for size-constrained searches, reducing disk usage for large scale deployments.
Scale
The search system scales to handle up to 10 million indexed files through intelligent engine selection. The system automatically uses Fast engine for small indexes and Semantic engine only when appropriate, preventing resource exhaustion from unnecessary embedding computation. This architecture enables enterprise deployments across large storage systems.