LLM powered reporting for digital ad campaigns
**Objective:** Our client invests over $500,000 daily in online advertising across networks like Meta and Google. They had several reporting teams analyzing data each day to optimize fund usage. The challenge was that data had to be downloaded as CSV files from each source, cleaned, and processed to create valuable reports for decision-makers. This lengthy process, which took 2-3 days, meant reports were always lagging and opportunities for better fund allocation were missed. Relia Software was engaged to develop a real-time reporting tool powered by a Large Language Model (LLM), integrating CPAS APIs from Meta, Google, etc. This tool provides a user-friendly chat interface for decision-makers to query and receive real-time insights regarding daily ad investment performance.
**Our Approach:** We developed an LLM-driven chat interface that connects with the Meta and Google CPAS Networks to produce on-demand reports for ad campaigns. Utilizing Langchain’s AI agents powered by OpenAI, the application can interpret natural language input and generate code for data analysis. Jupyter kernel gateway instances are employed to execute the code produced by the LLM. The outputs, which include text, tables, and graphs, are formatted by AI agents before being presented to users. The backend is designed to provide each user session with its own isolated Jupyter kernel gateway sandbox, ensuring data security and privacy. Additionally, the application includes a dashboard displaying preconfigured metrics derived from the latest campaign data and comes ready with prompt templates for automatic report generation at regular intervals.