How karthik14344/ai-incident-doctor works
AI Incident Doctor is a fault-injection and diagnosis project with two deliberately separate systems: KnowledgeAI, the patient application, and the Doctor, which diagnoses reported or alerted incidents. The user-facing path runs from the React interface through the gateway to document ingestion, retrieval, and model-backed answers. The operational path runs from monitoring alerts or user tickets through evidence collection, suspect-code retrieval, diagnosis, fix verification, and reporting. The README documents further capabilities whose internal wiring was not sampled; those relationships are cited to the README. Source excerpts and the index support the remaining relationships.
Main entry points: reasoner.py Root-cause reasoning · app/main.py Incident intake · app/main.py API gateway · app/main.py Document ingestion
Components 27
Patient interface
- React interface
patient/frontend/src/App.jsx
KnowledgeAI services
- API gateway
patient/api_gateway/app/main.py - Chat and document records
patient/api_gateway/app/db.py - Document ingestion
patient/ingestion_service/app/main.py - Chunking and embeddings
patient/ingestion_service/app/chunker.py - Vector store
patient/ingestion_service/app/vector_store.py - Context retrieval
patient/retrieval_service/app/main.py - Answer generation
patient/llm_service/app/main.py
Incident doctor
- Incident intake
doctor/app/main.py - Evidence collection
doctor/app/collectors.py - Evidence bundle
doctor/app/evidence.py - Suspect retrieval
doctor/app/retrieval.py - Search indexes
doctor/app/indexes.py - Root-cause reasoning
doctor/app/reasoner.py - Doctor model client
doctor/app/llm.py - Proposed code fix
doctor/app/fixes.py - Fix verification
doctor/app/verify.py - Incident report
doctor/app/report.py
Operations and evaluation
- Monitoring and alerts
monitoring/grafana/build_dashboard.py - Alert sink
alert_sink/sink.py - Fault injection
faults/run_fault.py - Patient evaluation
patient/evaluation/pipeline.py - Doctor evaluation
doctor/eval/run_eval.py
Other components
- User
- Ollama model service
- Ticket reporter
- Incident replay
Connections 26
- User → React interface: uses
- React interface → API gateway: calls API
- API gateway → Chat and document records: reads and writes
patient/api_gateway/app/main.py - API gateway → Document ingestion: dispatches processing
patient/api_gateway/app/main.py - Document ingestion → Chunking and embeddings: creates chunks
patient/ingestion_service/app/main.py - Document ingestion → Vector store: stores vectors
patient/ingestion_service/app/main.py - API gateway → Context retrieval: requests context
- Context retrieval → Vector store: queries vectors
patient/retrieval_service/app/main.py - API gateway → Answer generation: requests answer
- Answer generation → Ollama model service: generates tokens
patient/llm_service/app/main.py - Monitoring and alerts → Alert sink: sends alert
- Alert sink → Incident intake: forwards alert
- Ticket reporter → Incident intake: submits ticket
- Incident replay → Incident intake: starts replay
- Incident intake → Evidence collection: collects clues
doctor/app/main.py - Evidence collection → Evidence bundle: builds bundle
- Incident intake → Root-cause reasoning: diagnoses bundle
doctor/app/main.py - Root-cause reasoning → Suspect retrieval: retrieves context
doctor/app/reasoner.py - Suspect retrieval → Search indexes: uses indexes
doctor/app/retrieval.py - Root-cause reasoning → Doctor model client: requests reasoning
doctor/app/reasoner.py - Root-cause reasoning → Proposed code fix: proposes repair
doctor/app/reasoner.py - Root-cause reasoning → Fix verification: checks fix
doctor/app/reasoner.py - Incident intake → Incident report: writes report
doctor/app/main.py - Fault injection → User: disrupts patient
- Patient evaluation → Context retrieval: evaluates pipeline
patient/evaluation/pipeline.py - Doctor evaluation → Root-cause reasoning: evaluates diagnosis
doctor/eval/run_eval.py