Spaces:
Running
Running
| # InsightFlow AI - LangGraph Implementation Flow | |
| +---------------------+ +-------------------------+ +------------------------+ +-----------------+ +-----+ | |
| | | | | | | | | | | | |
| | planner_agent +----->+ execute_persona_tasks +----->+ synthesize_responses +----->+ present_results +----->+ END | | |
| | | | | | | | | | | | |
| +---------------------+ +-------------------------+ +------------------------+ +-----------------+ +-----+ | |
| Plan research Generate perspectives Combine perspectives Present to user | |
| - Examine query - Create persona instances - Collect all responses - Show synthesized view | |
| - Identify personas - Parallel API requests - Create unified view - Display individual | |
| - Structure approach - Handle timeouts - Structure insights perspectives if enabled | |
| Current Implementation Architecture: | |
| 1. Node: planner_agent | |
| - Input: user query and selected personas | |
| - Process: Plans approach (currently simplified) | |
| - Output: Updated state with research plan | |
| - Progress: 10% | |
| 2. Node: execute_persona_tasks | |
| - Input: State with query and personas | |
| - Process: Parallel generation of perspectives from each persona | |
| - Output: State with persona_responses dict | |
| - Progress: 40-80% (dynamically updates per persona) | |
| 3. Node: synthesize_responses | |
| - Input: State with all persona responses | |
| - Process: Creates unified view combining all perspectives | |
| - Output: State with synthesized_response | |
| - Progress: 80-95% | |
| 4. Node: present_results | |
| - Input: State with synthesized response and persona responses | |
| - Process: Formats and sends messages to user | |
| - Output: Final state ready to END | |
| - Progress: 95-100% | |
| Data Flow: | |
| - InsightFlowState object maintains the query state and responses | |
| - User-selected personas determine which reasoning approaches are used | |
| - Each persona operates independently but shares the same query | |
| - Final synthesis combines all perspectives into a unified view | |
| Features: | |
| - Parallel processing for faster responses | |
| - Timeout handling for reliability | |
| - Dynamic progress tracking for better UX | |
| - Direct mode bypass for speed when needed | |
| - Quick mode for reduced persona count | |
| - Streaming for real-time updates |