Bank M&A Ledger Reconciliation System using AutoGen Multi-Agent Simulation
When major banking institutions execute portfolio acquisitions (such as Santander integrating Webster Bank systems), merging legacy general ledgers and core deposit accounts takes months of manual reconciliation and auditing. Discrepancies in ledger naming conventions, chart-of-accounts mappings, and accrued interest logic often cause millions in post-merger accounting drift. In this challenge, you will implement a multi-agent conversational simulation using Microsoft AutoGen. Specialized agents representing the Acquiring Bank, Target Bank, and Independent Auditor will interact autonomously to verify balance sheet migration accuracy and resolve chart-of-accounts discrepancies.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an AutoGen multi-agent system to simulate and perform automated ledger reconciliation during bank acquisitions.
How work is evaluated
Evaluates reconciliation accuracy and agent consensus resolution.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
consensus_achieved
Verifies that all three agents reached explicit agreement
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
reconciliation_accuracy
Percentage of balance sheet items correctly mapped without residual error • target: 100 • range: 0-100
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Configure AutoGen ConversationalAgents with custom system messages representing distinct banking entities
Design structured dialogue rules and termination conditions for automated audit consensus
Map non-standard core banking account codes to unified chart-of-accounts structures
Output a validated reconciliation delta report with complete conversation history
Reference links and supporting material
Synthetic double-entry general ledger files representing two merging regional financial institutions.
Challenge setup
Evaluates reconciliation accuracy and agent consensus resolution.
Challenge input
JSON containing bank_a_ledger and bank_b_ledger arrays
AutoGen
Microsoft framework specializing in multi-agent conversational interaction.
Evaluated output
JSON containing unified_ledger, total_unmatched_delta, and audit_consensus_reached
- Verifies that all three agents reached explicit agreement
- Percentage of balance sheet items correctly mapped without residual error • target: 100 • range: 0-100
- Benchmark: FinAccounting-Reconciliation-Bench
- Reconciliation Accuracy target: 100
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- AutoGen · Required
- Microsoft AutoGen · Optional
Evaluation contract
- consensus_achieved · Weight 1
- reconciliation_accuracy · Weight 1
Metrics
- reconciliation_accuracy · Target 100
Benchmarks
- FinAccounting-Reconciliation-Bench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing bank_a_ledger and bank_b_ledger arrays
Output: JSON containing unified_ledger, total_unmatched_delta, and audit_consensus_reached
Execution harness: Python
Evaluation tools
- AutoGen · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.