Home
Projects
Tools
Architecture
Variations
Phoenix
Phase Status
Phase
Status
Components
Phase 0: Foundation Complete CLI, DataClient, ToolRegistry, ExecutionContext
Phase 1: Glass Box Complete PlanBuilder, AgenticEngine, BlueprintPanel, ExecutionTrace
Phase 2: Human-in-Loop Complete Pause/Resume/Cancel controls, InspectorPanel
Phase 3: Progressive Disclosure Complete Expandable blueprint details, tabbed inspector
Phase 4: Multi-Modal Output Complete ArtifactWorkspace, charts, PNG/CSV/JSON export
Architecture Layers
+------------------------------------------------------------------+
| AI_WQ Unified Interface |
| +---------------+ +---------------+ +---------------+ |
| | Project | | Model | | Notebook | |
| | Explorer | | Discovery | | Browser | |
| +---------------+ +---------------+ +---------------+ |
+------------------------------------------------------------------+
|
v
+------------------------------------------------------------------+
| ToolRegistry (~18 tools) |
| +--------------------------------------------------------------+ |
| | PROJECT TOOLS (2 per project) | |
| | - load_proj{1-6}_info | |
| | - list_proj{N}_notebooks/models/modules | |
| +--------------------------------------------------------------+ |
| | CROSS-PROJECT TOOLS | |
| | - compare_projects, find_similar_projects | |
| | - list_all_models, suggest_technique | |
| +--------------------------------------------------------------+ |
| | UTILITY TOOLS | |
| | - show_capabilities, get_project_info | |
| +--------------------------------------------------------------+ |
+------------------------------------------------------------------+
|
v
+------------------------------------------------------------------+
| ExecutionContext (3-Tier Cache) |
| step_results: Dict[int, Any] # Per-execution step outputs |
| model_cache: Dict[str, Any] # Loaded models |
| data_cache: Dict[str, Any] # Project data caching |
+------------------------------------------------------------------+
|
v
+------------------------------------------------------------------+
| AIDataClient |
| Source: ~/Python/WQ_AI/AI/ |
| - Proj1: Camera trap images, CNN models |
| - Proj2: Cassava dataset, pretrained/fine-tuned models |
| - Proj3: YOLO datasets, augmentation pipelines |
| - Proj4: Face datasets, MTCNN/FaceNet models |
| - Proj5: GAN training data, generator/discriminator |
| - Proj6: Diffusion prompts, pipeline configs |
+------------------------------------------------------------------+
Agentic Engine State Machine
+--------+
| IDLE |
+--------+
|
v
+-----------+
| PLANNING | <-- parse_query()
+-----------+
|
v
+---------------------+
| AWAITING_APPROVAL | <-- approve_plan()
+---------------------+
|
v
+---------+
| RUNNING | -----> COMPLETED
+---------+
| ^
v |
+--------+
| PAUSED | <-- pause() / resume()
+--------+
|
v
+-----------+
| CANCELLED |
+-----------+
File Structure
domains/AI_WQ/
+-- cli/
| +-- main.py # Phase 0: CLI interface (cmd.Cmd)
+-- cockpit/
| +-- ai_wq_cockpit.py # Phase 1-4: tkinter GUI
| +-- ui_components/
| +-- blueprint_panel.py # Plan preview
| +-- execution_trace.py # Real-time log
| +-- inspector_panel.py # Detail drill-down
| +-- artifact_workspace.py # Multi-modal output
+-- unified_agent/
| +-- data_client.py # AIDataClient
| +-- tool_registry.py # ~18 tools
| +-- execution_context.py # 3-tier caching
| +-- plan_builder.py # Query -> Plan
| +-- agentic_engine.py # State machine
+-- variations/
| +-- variation_1/ # 6 tools (Proj 1-2)
| +-- variation_2/ # 10 tools (+ Proj 3-4)
| +-- variation_3/ # 18 tools (+ Proj 5-6, cross-project)
+-- Proj1-6/ # Symlinks to source data
+-- DOMAIN_README.md
Plan Builder Pattern Priority
The plan builder uses regex pattern matching with the following priority:
Cross-project patterns - "compare", "similar", "suggest"Project-specific patterns - "project 1", "yolo", "gan", "diffusion"Exploration patterns - "list models", "list projects", "info N"Fallback - "explore", "overview"
More specific patterns must be matched first to ensure correct plan generation.
Example Query Flows
Flow 1: Project Exploration
User: "What's available in the face recognition project?"
-> load_proj4_info
-> list_proj4_modules
-> Result: MTCNN, InceptionResNet notebooks + Python modules
Flow 2: Cross-Project Analysis
User: "Compare CNNs vs YOLOv8 for object detection"
-> compare_projects(proj1, proj3)
-> suggest_technique("detection")
-> Result: Technique comparison with recommendations
Flow 3: Model Discovery
User: "What pre-trained models are available?"
-> list_all_models
-> Result: Models by project with paths
Flow 4: Technique Recommendation
User: "Best approach for image generation?"
-> suggest_technique("generation")
-> Result: GAN (Proj5) or Diffusion (Proj6) recommendations
Key Design Decisions
Lightweight Implementation
Focus on exploration and metadata; heavy ML ops use source notebooks directly
Source Reference
Data lives at ~/Python/WQ_AI/AI/, accessed via symlinks for separation
Tool Composition
Cross-project tools compose project-specific tools for complex queries
Educational Focus
Tools designed for learning and exploration, not production inference