How iamdineshreddy/adaptive-multi-agent-pr-review works
This is an asynchronous GitHub pull-request review system whose central contribution is selecting useful findings, not merely detecting possible issues. The main workflow runs from verified webhook ingestion through priority scheduling, multi-agent analysis, finding consolidation, and ARUM selection; developer feedback informs repository memory and later reviews. The graph preserves the dashboard and experiment harness as separate product capabilities. Some links rely on README architecture statements because the corresponding implementation files were not sampled.
Main entry points: service.py Review orchestrator · persistence.py Review persistence · adaptive ARUM selection · runner.py Experiment runner
Components 25
Webhook and queue
- Webhook handling
backend/app/webhooks/router.py - PR ingestion
backend/app/webhooks/ingest.py - Priority queue
backend/app/queue/tasks.py - Review store
backend/app/database/__init__.py
Review execution
- Review orchestrator
backend/app/orchestrator/service.py - Review agents
backend/app/agents - GitHub client
backend/app/github/client.py - Finding consolidation
backend/app/consolidation - Iterative review
backend/app/orchestrator/iteration.py - Review persistence
backend/app/orchestrator/persistence.py
Selection and learning
- ARUM selection
backend/app/adaptive - Scoring features
backend/app/adaptive/features.py - Repository memory
backend/app/adaptive/memory.py - Learned weights
backend/app/adaptive/weights.py - Historical RAG
backend/app/adaptive/rag.py - Developer feedback
backend/app/api/feedback.py
Product interface
- Review dashboard
frontend/src/App.tsx - Dashboard API client
frontend/src/services/api.ts - Backend API routes
backend/app/api/router.py - Dashboard data
backend/app/api/dashboard.py - Observability
backend/app/monitoring/prometheus.py
Experiment harness
- Experiment corpus
backend/app/experiments/corpus.py - Experiment modes
backend/app/experiments/modes.py - Experiment runner
backend/app/experiments/runner.py
Other components
- GitHub PR
Connections 26
- GitHub PR → Webhook handling: sends events
- Webhook handling → PR ingestion: validates and ingests
backend/app/webhooks/router.py - PR ingestion → Review store: persists review
backend/app/webhooks/ingest.py - Webhook handling → Priority queue: dispatches review
backend/app/webhooks/router.py - Priority queue → Review orchestrator: dispatches review
backend/app/queue/tasks.py - Review orchestrator → Review agents: runs agents
backend/app/orchestrator/service.py - Review orchestrator → Finding consolidation: consolidates findings
backend/app/orchestrator/service.py - Review orchestrator → ARUM selection: scores candidates
backend/app/orchestrator/service.py - ARUM selection → Scoring features: extracts features
backend/app/orchestrator/service.py - ARUM selection → Learned weights: loads weights
backend/app/orchestrator/service.py - Review orchestrator → Repository memory: refreshes memory
backend/app/orchestrator/service.py - Review orchestrator → Historical RAG: retrieves context
backend/app/orchestrator/service.py - Review orchestrator → Review persistence: records outcomes
backend/app/orchestrator/service.py - Review orchestrator → Iterative review: plans re-review
backend/app/orchestrator/service.py - Review orchestrator → GitHub client: fetches PR context
- Developer feedback → Review persistence: records feedback
- Review persistence → Repository memory: stores review memory
backend/app/orchestrator/persistence.py - Review dashboard → Dashboard API client: loads views
frontend/src/App.tsx - Dashboard API client → Backend API routes: requests data
frontend/src/services/api.ts - Backend API routes → Dashboard data: routes dashboard requests
backend/app/api/router.py - Dashboard data → Review persistence: reads review data
backend/app/api/dashboard.py - Observability → Review persistence: exports review metrics
backend/app/monitoring/worker_exporter.py - Experiment corpus → Experiment modes: supplies corpus
backend/app/experiments/runner.py - Experiment runner → Experiment corpus: loads corpus
backend/app/experiments/runner.py - Experiment runner → Experiment modes: runs modes
backend/app/experiments/runner.py - Experiment runner → ARUM selection: evaluates selection
backend/app/experiments/runner.py