How krishjain10/reconciliation-copilot works
Reconciliation Copilot is a Streamlit application for reconciling merchant ledger batches against bank settlement payouts. Its principal workflow loads and validates CSV inputs, deterministically calculates expected net payouts and classifies batches, then anonymizes unresolved records before generating guarded explanations and confidence scores. The dashboard presents resolved, unresolved, orphan, and audit views; AI output is advisory. The graph also includes the documented golden evaluation capability. Edges rely on the README or shown source/index references; unsampled UI and agent wiring is not inferred beyond those references.
Main entry points: app.py Dashboard Orchestrator · explain.py Explanation Agent · golden_set.py Golden Evaluation · engine.py Batch Matcher
Components 13
Reconciliation Workflow
- Dashboard Orchestrator
ui/app.py - Ledger And Settlement CSVs
data - CSV Loader
ingestion/loader.py - Batch Matcher
matching/engine.py
Privacy And Explanations
- PII Tokenizer
agent/anonymize.py - Explanation Agent
agent/explain.py - PII Guard
agent/pii_guard.py - Rule-Based LLM
ui/mock_llm.py
Dashboard
- Dashboard Views
ui/components.py - Dashboard Styling
ui/styles.py
Evaluation
- Golden Evaluation
eval/golden_set.py
Other components
- Finance Operator
- LLM Callable
Connections 13
- Finance Operator → Dashboard Orchestrator: runs reconciliation
README.md - Dashboard Orchestrator → Ledger And Settlement CSVs: accepts inputs
README.md - Dashboard Orchestrator → CSV Loader: loads CSVs
ui/app.py - Dashboard Orchestrator → Batch Matcher: matches batches
ui/app.py - Dashboard Orchestrator → PII Tokenizer: anonymizes records
ui/app.py - Dashboard Orchestrator → Explanation Agent: explains mismatches
ui/app.py - Dashboard Orchestrator → Rule-Based LLM: selects fallback
ui/app.py - Explanation Agent → PII Guard: guards prompts
agent/explain.py - Explanation Agent → LLM Callable: requests explanation
agent/explain.py - Dashboard Orchestrator → Dashboard Views: renders results
ui/app.py - Dashboard Orchestrator → Dashboard Styling: injects styles
ui/app.py - Dashboard Orchestrator → Golden Evaluation: runs evaluation
README.md - Golden Evaluation → Batch Matcher: evaluates matching
eval/golden_set.py