How rahulramachandran-labs/reconmind works
ReconMind is an incident copilot for diagnosing data-pipeline anomalies. A user initiates a scan or asks a question in the web app; the API runs a multi-agent investigation that gathers read-only warehouse evidence, checks domain-specific failure modes, retrieves runbooks and past incidents, and produces findings and a report. Serious or uncertain cases pause for human review. The app also exposes volume dashboards, incident history, verification, and run traces. Edges reflect README statements and the supplied source excerpts/index; unsampled implementation details are not assumed.
Main entry points: main.py FastAPI App · service.py Investigation Service · graph.py Agent Graph · knowledge.py Knowledge API
Components 26
User Experience
- ReconMind Web
frontend/src/app/page.tsx - Incident Views
frontend/src/app/incidents/page.tsx - Review Queue
frontend/src/app/review/page.tsx - Anomaly Verification
frontend/src/app/verify/page.tsx - Trace Views
frontend/src/app/traces/page.tsx
API and Runs
- FastAPI App
app/main.py - Agent API
app/api/agents.py - Knowledge API
app/api/knowledge.py - Dashboard API
app/api/dashboard.py - Investigation Service
app/agents/service.py
Investigation
- Agent Graph
app/agents/graph.py - Planner and Specialists
app/agents/specialists.py - Retail Recon Adapter
app/domain/retail_recon/adapter.py - Recon Checks
app/domain/retail_recon/checks.py - Human Sign-off
app/agents/review.py
Evidence and Retrieval
- MCP Tool Access
app/tools/mcp_toolbox.py - Warehouse Metadata MCP
mcp_servers/warehouse_metadata/server.py - Orchestration Metadata MCP
mcp_servers/orchestration_metadata/server.py - Hybrid Retrieval
app/retrieval/service.py - RAG Answering
app/rag/answer.py
State and Observability
- LLM Provider Chain
app/llm/providers.py - Run and Incident Store
app/agents/store - Chat Sessions
app/memory/sessions.py - PostgreSQL
app/db/models.py - Run Tracing
app/observability/tracer.py
Other components
- Analyst / Reviewer
Connections 30
- Analyst / Reviewer → ReconMind Web: uses
README.md - ReconMind Web → FastAPI App: requests
- Analyst / Reviewer → Incident Views: browses incidents
README.md - Analyst / Reviewer → Review Queue: reviews findings
README.md - Analyst / Reviewer → Anomaly Verification: verifies anomalies
README.md - Analyst / Reviewer → Trace Views: inspects runs
README.md - FastAPI App → Agent API: mounts router
app/main.py - FastAPI App → Knowledge API: mounts router
app/main.py - FastAPI App → Dashboard API: mounts router
app/main.py - Agent API → Investigation Service: invokes
app/api/agents.py - Knowledge API → Hybrid Retrieval: searches
app/api/knowledge.py - Knowledge API → RAG Answering: answers with
app/api/knowledge.py - Investigation Service → Agent Graph: runs
app/agents/service.py - Agent Graph → Planner and Specialists: dispatches
app/agents/graph.py - Planner and Specialists → Retail Recon Adapter: uses domain contract
app/agents/specialists.py - Retail Recon Adapter → Recon Checks: performs checks
- Investigation Service → Human Sign-off: pauses for review
app/agents/service.py - Agent Graph → MCP Tool Access: gathers evidence
README.md - MCP Tool Access → Warehouse Metadata MCP: calls tools
README.md - MCP Tool Access → Orchestration Metadata MCP: calls tools
README.md - Agent Graph → Hybrid Retrieval: retrieves context
- RAG Answering → Hybrid Retrieval: searches
app/rag/answer.py - RAG Answering → LLM Provider Chain: generates answer
app/rag/answer.py - Planner and Specialists → LLM Provider Chain: analyzes findings
app/agents/specialists.py - FastAPI App → LLM Provider Chain: initializes
app/main.py - Investigation Service → Run and Incident Store: persists run data
- FastAPI App → Chat Sessions: selects session store
app/main.py - Chat Sessions → PostgreSQL: stores sessions
app/memory/sessions.py - Investigation Service → Run Tracing: records run steps
app/agents/service.py - Run Tracing → PostgreSQL: writes steps
app/observability/tracer.py