Context and constraints
Bancor is Banco de CĂłrdoba, a provincial bank. Its digital platform is a regulated production environment: anything added to it has to meet the bank's security, compliance and audit requirements before it reaches an employee or a customer.
Rather than running experiments beside that platform, the bank set up an AI engineering capability inside it. Caramel provides that team: senior engineers embedded with the bank's own people, working on the bank's systems and data.
The architecture decision
Agents and LLM tools are built inside the platform and connected to the bank's knowledge and systems, not as a separate assistant on the side. Retrieval-augmented generation (RAG) grounds each answer in the bank's own documentation, and the components run in production, not in a sandbox.
The reasoning is the one we apply everywhere: automation and decision support only pay off when they act on the systems the organisation already runs. An assistant beside the platform adds a channel; an engineering team inside it adds capabilities.
What the client got
- Critical processes automated inside the bank's platform.
- RAG running in banking production, grounded in the bank's own documentation.
Stack
- LLM Integration
- AI Agents
- Python
- NLP
- Azure