Building AI Applications in the Enterprise Part 1 | WSO2Con Barcelona 2025



AI Summary

Summary of Video: Generative AI Applications

Introduction

  • Overview of practical use cases of Generative AI (GenAI).
  • Discussion of the evolution from simple to complex use cases and integration knowledge.

Key Concepts

  • Generative AI (GenAI): A subset of AI that generates new content (text, images, audio) based on existing patterns.
  • Large Language Models (LLMs): Specialize in text generation, e.g., the impact of the “Attention is All You Need” paper.

Practical Use Case: AI Assistants

  • Examples of AI features integrated into applications (e.g., Q&A sessions, session advisor, expert finder).
  • Implementation Steps:
    • Design user interactions and prompts.
    • Develop an HTTP service for chat functionality.
    • Use OpenAI’s API for model integration and responses.

Challenges & Solutions

  • Hallucination Problem: Risk of generating incorrect answers due to training data limitations.
    • Suggested solutions:
      • Fine-tuning models (complex and time-consuming).
      • In-context learning: Incorporate real-time data via prompts.

Advanced Techniques

  • Retrieval Augmented Generation (RAG): Enhances GenAI applications by filtering relevant data before generation.
  • Use of vector databases for data retrieval and embedding for efficient information processing.

AI Agent Integration

  • Concept of AI agents that dynamically choose tools to execute tasks based on user queries.
  • Importance of integrating various data sources to enhance application capabilities (e.g., fetching speaker information and session details).

Conclusion

  • Successful GenAI applications require accuracy, acceptable latency, scalability, secure data access, and the ability to avoid misuse.
  • Integration approaches should balance complexity with risk management, ensuring tools are agent-compatible (e.g., restructuring APIs for better agent usability).