Data & Innovation • Financial Services • AWS
GenAI and information confidentiality: Development of a secure chatbot with document understanding
Integration of Generative Artificial Intelligence in controlled environments, focused on privacy, efficiency, and user experience.
About the client
A major Argentine bank that combines a deep-rooted tradition with a modern vision of the future. With a strong focus on closeness and innovation, it develops solutions designed to support both individuals and companies in their daily challenges, offering a banking experience that is agile, human, and efficient. Its essence lies in building long-term relationships of trust, understanding the real needs of each client.
Driven by digital transformation and a commitment to the community, it seeks to be much more than just a bank: an ally in decision-making, a driver of growth, and a bridge to new opportunities. Through technology, personalized advisory services, and an active presence throughout the country, it continues to strengthen its value proposition in a constantly evolving environment.
Needs
The bank required a Proof of Concept (PoC) for Generative Artificial Intelligence (GenAI) to implement an internal chatbot that could query PDF documents, with a strong focus on security and privacy. The goal was to evaluate these technologies within a controlled environment and avoid the use of public tools that could compromise confidential information.
Solution
Nubiral designed a Profile as a Service environment to develop a chatbot capable of dynamically uploading PDF files by users.
The solution enabled queries on the content of each document and generated assisted text, with memory retained for up to five interactions.
In addition, user authentication, usage traceability, and an analytics layer for monitoring consumption were incorporated.
Results
A robust conversational interface was developed, supported by an AI model that interacts with an OpenSearch index.
The chatbot evaluates each query to determine whether it should access the index or respond based on the conversation history, which remains active at all times.
All chats are stored in S3, and tokens are recorded in DynamoDB for usage control.
This implementation allowed the bank to successfully explore GenAI capabilities in a secure environment tailored to its operational needs.
