Placeholder image

Agentic AI for Financial Insights

| ChanJin Park |

AI CFO

YouTube

Anchor

Spotify

Apple

Episode #237

Introduction

In episode 237 of our SAP on Azure video podcast we talk about again about SAP and AI. When I presented at the DSAG Technologydays a few weeks I started to talk about the Joule & Copilot integration. While I think this is going to be a great integration of the UI for AIs, I can see a lot of other activities happening when it comes to SAP and AI. Week by week our colleauge CJ is publishing great articles on LinkedIn about AI can work with SAP, from Agentic AI Workflow for SAP to Leveraging Semantic Kernel + Azure AI Agents + Code Interpreter for CFO Office. I am glad to have him back on our show to talk more about this integration.

Find all the links mentioned here: https://www.saponazurepodcast.de/episode237

Reach out to us for any feedback / questions:

#Microsoft #SAP #Azure #SAPonAzure #AI #Agentic #SemanticKernel

Summary created by AI

  • Holger introduced the episode and welcomed CJ to discuss AI integration with SAP. Holger praised CJ’s work on LinkedIn and mentioned the topics to be covered.
  • CJ’s Role:
  • CJ introduced themselves as part of the SAP on Azure team, helping customers and partners with AI adoption.
  • Agentic AI:
  • CJ explained the concept of agentic AI, where individual AI agents perform specific tasks with human supervision.
    • Agentic AI Concept: CJ explained the concept of agentic AI, where individual AI agents perform specific tasks with human supervision. This involves using AI agents to handle tasks previously managed by RPA, integrating LLM-based APIs into copilot functionalities, and enabling human supervisors to oversee AI agents.
    • Single AI Agent: CJ described the stage of a single AI agent, where a user interacts with a chatbot through a single AI agent supervised by a human. This setup allows for more efficient task management and integration of AI capabilities.
    • Multi-Agent System: CJ discussed the evolution from single AI agents to multi-agent systems, where multiple AI agents handle different tasks within an organization, such as expense management, compliance, and cash flow forecasting, orchestrated by a CFO.
  • Azure AI Foundry SDK:
  • CJ discussed the continuous updates to the Azure AI Foundry SDK and the availability of various models for integration.
    • SDK Updates: CJ highlighted the continuous updates to the Azure AI Foundry SDK, mentioning the availability of various models for integration, including over 1600 models in the model catalog, such as GPT-4.
    • Model Catalog: Holger added that the model catalog includes a wide range of models, providing users with a significant choice for their AI integration needs.
  • Multi-Agent System:
  • CJ described the idea of using multiple agents for different tasks within the CFO office, such as expense management, compliance, and cash flow forecasting.
    • Multi-Agent Concept: CJ described the concept of using multiple agents for different tasks within the CFO office, such as expense management, compliance, and cash flow forecasting. Each agent handles specific tasks, and they interact with each other to provide comprehensive insights and decisions.
    • Agent Roles: Holger summarized the roles of different agents, including expense agent, compliance agent, forecasting agent, and others, each responsible for specific tasks within the CFO office.
    • Customer Use Cases: CJ shared insights from discussions with customers and partners, highlighting the practical applications of multi-agent systems in managing financial data, compliance, and forecasting within organizations.
  • Data Replication:
  • CJ mentioned the importance of replicating SAP data into fabric using connectors and the benefits of having data available for consumption.
    • Data Replication: CJ emphasized the importance of replicating SAP data into fabric using connectors, such as table connectors and HANA connectors, to make data available for consumption and integration with AI agents.
    • Connector Benefits: CJ highlighted the benefits of using connectors to replicate data, enabling seamless integration and accessibility of SAP data within fabric for various AI applications.
  • Fabric Integration:
  • CJ highlighted the recent announcement of integrating fabric with Azure AI agent directly, allowing seamless data consumption.
    • Integration Announcement: CJ announced the recent integration of fabric with Azure AI agent directly, allowing seamless data consumption and interaction between fabric data and AI agents.
    • Data Agent: CJ explained the concept of a data agent within fabric, which enables direct integration and consumption of fabric data by Azure AI agents, enhancing the efficiency and capabilities of AI applications.
    • Customer Benefits: CJ shared the benefits of this integration for customers, including improved data accessibility, streamlined processes, and enhanced AI capabilities for managing and analyzing SAP data.
  • Semantic Kernel:
  • CJ explained the role of the semantic kernel in orchestrating business processes and automating tasks using plugins.
    • Semantic Kernel Role: CJ explained the role of the semantic kernel in orchestrating business processes and automating tasks using plugins. The semantic kernel supports various languages and enables modular extensions for business process automation.
    • Plugin Functionality: CJ described how plugins within the semantic kernel are used to define custom functions for tasks like data grouping, aggregation, and analysis, facilitating efficient business process automation.
    • Orchestration Tool: CJ highlighted the semantic kernel as a fully supported production-level orchestration tool by Microsoft, enabling the automation and extension of business processes through plugins.
  • Custom Functions:
  • CJ demonstrated how custom functions are defined within plugins to perform tasks like data grouping and aggregation.
    • Function Definition: CJ demonstrated how custom functions are defined within plugins to perform tasks like data grouping and aggregation. These functions are written in code and integrated into the semantic kernel for execution.
    • Data Grouping: CJ provided an example of a custom function for data grouping, explaining how it reads table data, performs grouping and aggregation, and returns the processed data for further analysis.
    • Plugin Integration: CJ explained the integration of custom functions within plugins, enabling the semantic kernel to execute specific tasks based on user prompts and predefined instructions.
  • Planner:
  • CJ described the planner’s role in routing user prompts to the appropriate agents and executing tasks based on reasoning.
    • Planner Role: CJ described the planner’s role in routing user prompts to the appropriate agents and executing tasks based on reasoning. The planner reads user prompts, decides which agent should handle the task, and orchestrates the execution process.
    • Task Execution: CJ explained how the planner routes user prompts to the appropriate agents, ensuring that tasks are executed efficiently and accurately based on the predefined functions and reasoning.
    • Orchestration Process: CJ highlighted the orchestration process managed by the planner, which involves reading user prompts, routing tasks to agents, and executing functions to achieve the desired outcomes.
  • Code Interpreter:
  • CJ showed how the code interpreter reads and processes data, executing user prompts and returning results.
    • Code Interpreter Function: CJ demonstrated how the code interpreter reads and processes data, executing user prompts and returning results. The code interpreter handles tasks like reading CSV files, joining tables, and aggregating data.
    • Data Processing: CJ explained the data processing capabilities of the code interpreter, including reading table data, performing calculations, and returning processed results based on user prompts.
    • Execution Example: CJ provided an example of the code interpreter executing a user prompt, showing the step-by-step process of reading data, performing reasoning, and returning the final output.
  • Demo:
  • CJ provided a live demo of the system, showing how user prompts are processed and results are generated.
    • Live Demo: CJ provided a live demo of the system, showing how user prompts are processed and results are generated. The demo included examples of user prompts, the execution process, and the final output generated by the code interpreter.
    • Execution Flow: CJ demonstrated the execution flow of user prompts, including the invocation of Azure AI agents, the routing of tasks by the planner, and the processing of data by the code interpreter.
    • Result Generation: CJ showed how the system generates results based on user prompts, highlighting the efficiency and accuracy of the multi-agent system in handling complex tasks.

Links

Video index

Based on and using Hugo Cookbook Theme, Font Awesome Icons and Bulma CSS. | Thanks to Recipes4AJ to get me started. | Contact me via