Summary

Implement a comprehensive dashboard architecture allowing users to create custom analytical views of LLM observability data. This includes a widget configuration system, ClickHouse query transformer, and flexible data visualization capabilities.

Key Components

• Widget Configuration: Support for time series, aggregations, tables with dynamic filtering
• Query Transformer: Convert widget configs to optimized ClickHouse SQL with proper joins
• Dashboard Management: Save/share configurations with real-time updates

Implementation Scope

1. Core Infrastructure: Query transformer with column registry and comprehensive tests
2. API Layer: Dashboard CRUD endpoints with user permissions
3. Frontend: React-based builder with drag-and-drop interface

Acceptance Criteria

• Users can create custom dashboards with multiple widget types
• Real-time data from traces, observations, and scores tables
• Optimized ClickHouse queries with >95% test coverage
• Complete documentation and performance benchmarks

PR Connection

This issue will be addressed through the implementation of:

• ClickHouse Query Transformer (ClickHouseQueryTransformer.ts)
• Column Registry system (clickHouseColumnRegistry.ts)
• Widget configuration schemas and validation
• Comprehensive test suite with real data factories

Labels: enhancement, dashboard, clickhouse, analytics

Additional information

No response