Data & Innovation • Financial Services • Microsoft
GenAI solution with updatable embeddings and automated deployment on Azure
Solution based on Azure OpenAI with a document ingestion pipeline, automatic vectorization, API-based orchestration, and a conversational frontend through Google Chat.
About the client
A leading Mexican financial company specializing in credit solutions designed specifically for retirees and pensioners. With a focus on technological innovation, it leverages artificial intelligence and data analytics to provide personalized, accessible, and reliable financial products.
It has a broad network of commercial advisors distributed throughout the country, enabling a strong national presence and offering close, efficient customer service. Its commitment to continuous improvement and operational excellence ensures agile processes and a satisfying user experience.
The company also stands out for its organizational culture oriented toward employee well-being, having been recognized in workplace environment rankings within the financial sector. Its main purpose is to support older adults so they can enjoy their retirement years with financial backing and comprehensive advisory services.
Needs
The company sought to streamline access to information contained in PDF documents stored in its internal repository.
The goal was to provide internal users with a conversational tool capable of answering natural language queries about the content of these documents, thus enhancing operational efficiency and reducing reliance on manual processes.
Solution
Nubiral designed and implemented a Generative AI solution deployed on Azure.
The proposal included the use of OpenAI models (GPT and ADA) through Azure AI, enabling conversational interaction with documents.
Two possible user interfaces were offered, including Google Chat, and Infrastructure as Code (IaC) was applied for efficient implementation. In addition, mechanisms for periodic updating of embeddings were incorporated to keep the information current.
Results
The solution enables smooth and automated interaction with internal documents, facilitating information access without the need for complex ETL processes.
A monitoring system was implemented to provide visibility into tool usage, identifying the number of queries and the most accessed documents.
This improves decision-making and optimizes users’ daily work.
