How srikrishnamalisetti1106/azure-reconciliation works
The documented product reconciles processor payments against ledger entries, reports preliminary and final KPIs, and provides a read-only assistant to investigate exceptions. The sampled runtime confirms synthetic data generation and deterministic, uniqueness-checked matching with exception classification. The README describes additional pipeline and assistant stages, but their implementation files are absent from the supplied tree, so those nodes have no source path; their documented connections cite the README.
Main entry points: matching.py Versioned matching · generate.py Synthetic data generator
Components 14
Reconciliation pipeline
- Synthetic data generator
data/synthetic/generate.py - Bronze raw data
- Silver quality gates
- Versioned matching
reconciliation/matching.py - Gold results
Reporting
- T+1 / T+3 KPIs
- Dashboard
Investigation assistant
- FastAPI
- LangGraph agent
- Langfuse traces
Assistant data and search
- Read-only MCP tools
- Rule and KPI search
Other components
- Analyst
- Claude
Connections 14
- Synthetic data generator → Bronze raw data: supplies CSVs
README.md - Bronze raw data → Silver quality gates: types and gates
README.md - Silver quality gates → Versioned matching: provides records
README.md - Versioned matching → Gold results: writes results
README.md - Gold results → T+1 / T+3 KPIs: feeds snapshots
README.md - T+1 / T+3 KPIs → Dashboard: reports rates
README.md - Analyst → FastAPI: asks questions
README.md - FastAPI → LangGraph agent: routes requests
README.md - LangGraph agent → Claude: consults
README.md - Claude → LangGraph agent: returns response
README.md - LangGraph agent → Read-only MCP tools: invokes tools
README.md - LangGraph agent → Rule and KPI search: searches docs
README.md - LangGraph agent → Langfuse traces: emits traces
README.md - Gold results → Read-only MCP tools: read-only copy
README.md