NPL Case Review Agent
A single cockpit where the case manager asks the file questions and receives structured summaries for recovery decisions.
The business process
Digitisation of the full NPL portfolio analysis cycle: automatic file classification, structured extraction and generation of critical summaries for recovery valuation.
What the agent reads
- Massive ingestion through integration with the servicer's systems
- 26 core document categories: loan contracts, CTU appraisals, cadastral extracts, court documents, identity documents
What it produces
- Hundreds of datapoints grouped by logical area: identities of debtors and guarantors, exposure, collateral, judicial status
- Thematic summary sheets per case
How it works
Quick Actions let the manager query the file in natural language, for instance the security value of a given property. Targeted retrieval uses classification metadata to increase precision, and generated answers are compared with already extracted structured data, flagging any conflict.
Runs on
- Semantic knowledge graph
- GraphRAG & traversal queries
- Domain-tuned SLM/LLM/VLM serving
- Human-in-the-loop review
Related agents
Agents that share the same industry or the same capabilities.
Secured NPL Data Acquisition Agent
Reads 130 million pages of loan and property files and extracts 330 datapoints per case for due diligence.
Unsecured NPL Classification Agent
Classifies 3.7 million documents and pinpoints statute-of-limitation dates for unsecured portfolios.
NPL Payment Proposal Due Diligence Agent
Verifies every repayment plan against the customer database before it is sent to the debtor.