How parva-28/bob-ai-hackathon-duckduckgo works
YieldGuard is a semiconductor-yield investigation assistant: engineers inspect fleet and lot data, gather wafer-map, sensor, historical-case, and telemetry evidence, then review ranked root-cause hypotheses and containment actions. The web console fronts a FastAPI layer over MCP tools; IBM Bob can invoke those tools directly. The map preserves the documented vision, anomaly, planned-lot risk, conformal-abstention, telemetry, and playbook capabilities. Some UI-to-API and model-to-reasoning wiring is not shown in the excerpts, so those links are omitted or marked unverified.
Main entry points: server.py MCP Tool Server · main.py FastAPI Backend · adapters.py Tool Adapters · page.tsx Analyst Console
Components 23
Analyst Console
- Analyst Console
src/web/app/overview/page.tsx - Lot Investigation
src/web/app/investigation/page.tsx - Lot Analysis
src/web/app/lot-analysis/page.tsx - Batch Risk
src/web/app/batch-risk/page.tsx - Equipment View
src/web/app/equipment/page.tsx - Pipeline Trace
src/web/app/pipeline/page.tsx - Action Playbook
src/web/app/playbook/page.tsx
Tool Services
- FastAPI Backend
src/api/main.py - Input Validation
src/api/validate.py - MCP Tool Server
src/mcp_server/server.py - Tool Adapters
src/mcp_server/adapters.py - Parameter Mapping
src/mcp_server/param_map.py - Provenance
src/mcp_server/provenance.py - Lot Sensors
src/mcp_server/lot_sensors.py
Evidence Stores
- Case and Telemetry Stores
src/mcp_server/stores.py
ML Analytics
- Wafer Classifier
src/models/vision/classifier.py - Wafer CNN
src/models/vision/model.py - Sensor and Batch Risk
src/models/tabular/anomaly.py - CMP Prediction
src/models/cmp/predict.py - Conformal Abstention
src/models/cmp/abstain.py
Reasoning
- Root-Cause Reasoning
src/reasoning/reasoning.py
Other components
- Yield Engineer
- IBM Bob
Connections 23
- Yield Engineer → Analyst Console: inspects lots
src/web/app/overview/page.tsx - Yield Engineer → Lot Investigation: investigates lot
- Yield Engineer → Action Playbook: dispatches actions
src/web/app/playbook/page.tsx - Analyst Console → FastAPI Backend: requests data
- Lot Investigation → FastAPI Backend: requests analysis
- FastAPI Backend → MCP Tool Server: uses tool functions
src/api/main.py - FastAPI Backend → Input Validation: validates inputs
src/api/main.py - IBM Bob → MCP Tool Server: calls tools
README.md - MCP Tool Server → Tool Adapters: dispatches tools
src/mcp_server/server.py - MCP Tool Server → Parameter Mapping: maps parameters
src/mcp_server/server.py - MCP Tool Server → Provenance: records provenance
src/mcp_server/server.py - MCP Tool Server → Lot Sensors: uses lot sensors
src/mcp_server/server.py - Tool Adapters → Case and Telemetry Stores: uses stores
src/mcp_server/adapters.py - Tool Adapters → Wafer Classifier: classifies maps
src/mcp_server/adapters.py - Tool Adapters → CMP Prediction: predicts CMP
src/mcp_server/adapters.py - Tool Adapters → Sensor and Batch Risk: scores sensors
- Wafer Classifier → Wafer CNN: runs model
src/models/vision/classifier.py - CMP Prediction → Conformal Abstention: applies gates
src/models/cmp/predict.py - MCP Tool Server → Root-Cause Reasoning: ranks hypotheses
- Sensor and Batch Risk → Case and Telemetry Stores: matches historical cases
- Equipment View → FastAPI Backend: loads telemetry
- Batch Risk → FastAPI Backend: checks planned lots
- Pipeline Trace → FastAPI Backend: runs traced analysis