SmartAgent

Challenge

Itaú's business teams needed technical staff for every data query. Every question about the Data Lake went through the data or technology teams, so decisions waited in a queue and analysts spent their time handling requests instead of analyzing.

Information was fragmented across multiple tables, interpretation varied between teams, and decision-makers had no real-time visibility.

On top of that came a condition specific to banking: any generative AI solution on this data had to operate within the controls of an institution regulated by the BCU (Central Bank of Uruguay).

Our solution

Promtior designed and built SmartAgent, a multi-agent conversational assistant that answers business questions directly on the bank's governed data. No SQL and no intermediaries.

The architecture combines several language models, each assigned to specific functions within the flow. Among them is Claude, Anthropic's model, available through Amazon Bedrock, which handles specific tasks in the conversational processing. Everything runs inside the bank's own AWS environment.

The bank's controls were part of the design from the start:

• Every query is authenticated against the corporate Active Directory, so it always carries the user's real identity.
• The assistant's access follows the institution's data governance and retention policies.
• Everything runs on the bank's existing AWS infrastructure.

SmartAgent enables

• Real-time queries on customers, products, and profitability.
• Natural-language explanations of patterns, causes, and anomalies.
• Guided data exploration to uncover opportunities.
• Standardized decision logic across the bank.

In production

SmartAgent is in production at Banco Itaú Uruguay. The rollout was staged: in Q1 2026 the bank's data analysts started using it, and in Q3 2026 it opened to internal business areas. Validating first with a technical audience made it possible to open it to the business with confidence.