1259-UC Production Data Access Architecture for AI Agents Using a Proxy Layer Approach
Location
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Document Type
Event
Start Date
20-11-2026 2:00 PM
Description
Organizations increasingly rely on data to support business operations, evaluate performance, and make informed decisions. Traditionally, accessing and analyzing organizational data has required users to understand database structures, write SQL queries, or rely on specialized reporting tools and technical personnel. However, the development of AI and LLMs have offered possibilities for making data interaction possible as users no longer need to have specific tools to write queries and reports. They can now ask their questions in everyday language and receive meaningful responses. This transition provides the opportunity for AI-based reporting agents to serve as conversation interfaces to an organization’s data. Therefore, AI-powered reporting agents make accessing organizational information easier and lower the difficulty of traditional reporting methods. The purpose of this project is to design an AI-driven reporting agent that enables employees and managers to get the relevant organizational information by making use of conversational interface. In this proof of concept, there is a Gradio-based front-end, Python backend, and OpenAI’s GPT-OSS-120B model through the Groq Cloud API, and a PostgreSQL database. A primary consideration throughout development is protecting the database from unrestricted AI access. If the AI agent is given direct access to the database, it would be possible to expose sensitive information and create opportunities for unauthorized retrieval of data and unintended operations. To address these risks, the project heavily focuses on an intermediary API proxy that will include various role-based access controls and limitations on database operations. This approach is intended to ensure that requests made by the AI model to the database are properly validated and limited to available functionality.
1259-UC Production Data Access Architecture for AI Agents Using a Proxy Layer Approach
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Organizations increasingly rely on data to support business operations, evaluate performance, and make informed decisions. Traditionally, accessing and analyzing organizational data has required users to understand database structures, write SQL queries, or rely on specialized reporting tools and technical personnel. However, the development of AI and LLMs have offered possibilities for making data interaction possible as users no longer need to have specific tools to write queries and reports. They can now ask their questions in everyday language and receive meaningful responses. This transition provides the opportunity for AI-based reporting agents to serve as conversation interfaces to an organization’s data. Therefore, AI-powered reporting agents make accessing organizational information easier and lower the difficulty of traditional reporting methods. The purpose of this project is to design an AI-driven reporting agent that enables employees and managers to get the relevant organizational information by making use of conversational interface. In this proof of concept, there is a Gradio-based front-end, Python backend, and OpenAI’s GPT-OSS-120B model through the Groq Cloud API, and a PostgreSQL database. A primary consideration throughout development is protecting the database from unrestricted AI access. If the AI agent is given direct access to the database, it would be possible to expose sensitive information and create opportunities for unauthorized retrieval of data and unintended operations. To address these risks, the project heavily focuses on an intermediary API proxy that will include various role-based access controls and limitations on database operations. This approach is intended to ensure that requests made by the AI model to the database are properly validated and limited to available functionality.