Enterprise automation
Vera: Agentic AI for Enterprise Financial Reconciliation & Validation
A fully local Qwen-based Agent Validator that parses VAT outgoing listings and Coretax data against ERP records, achieving 97%+ accuracy across 14,000+ documents for the Tunas Rent finance team.
- accuracy
- 97%+over 14,000+ documents
- branches
- 159Tunas Rent operations
- docs/month
- 2.5K–3Kproduction throughput
01Problem
Finance reconciliation required manual cross-checking of VAT outgoing listings and Coretax raw data against ERP records. Errors were frequent and turnaround was slow across 159 branches and an 8,000+ vehicle fleet.
02Approach
- Built a fully local Qwen-based Agent Validator that parses VAT & Coretax sheets and reconciles them against internal ERP records.
- Tuned the validator for Indonesian financial-document structure and edge cases.
- Deployed the system on the internal company network with per-document processing time of 1–5 minutes.
03Contribution
- Designed and implemented the local Qwen agent pipeline end-to-end.
- Validated accuracy across a 14,000+ document holdout set before production rollout.
- Coordinated deployment within Tunas Group internal infrastructure.
04Outcome
- 97%+ accuracy across 14,000+ reconciliation documents.
- Production throughput of 2,500–3,000 documents per month across 159 branches.
- Replaced a slow manual workflow with a fully local, GPU-accelerated agent pipeline.
05Limitations
- The system is deployed on the internal company network and is not publicly accessible.
- Source code and document layouts are confidential to PT. Tunas Ridean Tbk.