How advait27/eadip works
EADIP is an enterprise decision-intelligence platform: users submit business questions, receive a live, evidence-grounded investigation, and get an independently verified executive brief. The principal flow spans question planning and execution, document and graph evidence, governed analytics, verification, and reporting; side-effecting tool actions pause for human approval. The README documents additional governance and integration capabilities, but excerpts do not show every route or dependency, so uncertain wiring is omitted.
Main entry points: service.py Run orchestrator · app.py API gateway · pipeline.py Ingestion pipeline · retrieval Hybrid retrieval
Components 23
API and identity
- API gateway
src/eadip/gateway/app.py - Run endpoints
src/eadip/gateway/routes/runs.py - Search endpoint
src/eadip/gateway/routes/search.py - Identity and access
src/eadip/security
Evidence and retrieval
- Ingestion pipeline
src/eadip/ingestion/pipeline.py - Hybrid retrieval
src/eadip/retrieval - GraphRAG evidence
src/eadip/graph/service.py - Vector store
src/eadip/adapters/qdrant_store.py - Document repository
src/eadip/adapters/postgres_document_repository.py
Analysis and reporting
- Governed analytics
src/eadip/analytics/service.py - SQL safety gate
src/eadip/analytics/sql_validator.py - Analytics warehouse
src/eadip/adapters/postgres_warehouse.py - Verification and brief
src/eadip/verification/service.py
Investigation execution
- Run orchestrator
src/eadip/orchestrator/service.py - Planning agents
src/eadip/agents - Step executors
src/eadip/orchestrator/executors.py - Run checkpoints
src/eadip/orchestrator/checkpoint.py
Governance and integrations
- Human approval
src/eadip/approval/service.py - MCP tool access
src/eadip/mcp - Audit logging
src/eadip/security/audit.py
Other components
- Business user
- LLM provider
- Approver
Connections 23
- Business user → API gateway: submits question
- API gateway → Identity and access: authenticates requests
- API gateway → Run endpoints: routes run requests
- API gateway → Search endpoint: routes searches
- Run endpoints → Run orchestrator: starts or streams runs
- Search endpoint → Hybrid retrieval: requests evidence
- Ingestion pipeline → Vector store: upserts chunks
- Ingestion pipeline → Document repository: records lineage
- Ingestion pipeline → Audit logging: records redactions
- Hybrid retrieval → Vector store: searches vectors
- Hybrid retrieval → GraphRAG evidence: merges graph evidence
- Run orchestrator → Planning agents: interprets and plans
- Run orchestrator → Step executors: dispatches steps
- Run orchestrator → Run checkpoints: saves run state
- Run orchestrator → Verification and brief: submits findings
- Run orchestrator → Human approval: gates proposed actions
- Human approval → Run checkpoints: updates run state
- Approver → Human approval: decides actions
- Step executors → MCP tool access: runs tool steps
- Governed analytics → SQL safety gate: validates queries
- Governed analytics → Analytics warehouse: executes safe SQL
- Planning agents → LLM provider: requests completions
- Verification and brief → LLM provider: requests recommendations