Embarking on your Generative AI Journey with Google | Pritam Sahoo | TEDxDSATM

Generative AI and Agentic AI: Solving Data Fragmentation and Enhancing Productivity

Key Concepts:

• Generative AI (GenAI)
• Agentic AI
• Data Fragmentation/Data Silos
• LLMs (Large Language Models)
• Indexing
• SAS (Software as a Service)
• Access Control Lists (ACLs)
• Multimodal Data
• Personalization
• Gemini
• NotebookLM

1. The Evolution of AI and the Problem of Data Fragmentation:

• The speaker positions the current state as "phase two" of generative AI, emphasizing its disruptive impact across industries.
• Enterprises are eager to leverage GenAI for business problem-solving, particularly in the realm of productivity and creativity.
• Past technologies (1980s-2000s) were fragmented, while more recent technologies (2010s-2020s) have progressed seamlessly.
• A major challenge remains: technologies struggle to communicate with each other, leading to data fragmentation and silos.
• Enterprises often deal with 6-8 disparate systems to access data, impacting productivity and wasting time on internal document searches.

2. The Overwhelming Pace of Technological Evolution:

• The rapid evolution of technology (AI, cloud platforms, software) forces organizations to adapt to stay competitive.
• Trends like reskilling in AI, digital transformation, and demand for personalized customer experiences drive the need for new strategies.
• Search, GenAI, and agents are key technologies, but their integration is often flawed.
• The speaker highlights the potential for uncontrolled agent proliferation, management challenges, and security risks (malicious content).
• These issues introduce risk, cost, and complexity.

3. Agentic AI: A Horizontal Solution for Various Industries:

• Agentic AI caters to various industries (healthcare, finance, insurance, public sector) with horizontal use cases for line-of-business users.
• The speaker, with an architect background, supports sales teams in addressing customer requirements.
• Example: Sales representatives waste time searching through lengthy tenders (800-1000 pages) for relevant information.
• Preparing for customer meetings requires accessing data from multiple systems (CRM, billing, marketing, Jira, Slack).
• Agentic AI can consolidate this data, providing intelligence and citations to ensure informed and productive customer interactions.
• This approach offers "Google-like search for your enterprise data," contextualized and personalized.

4. Multimodal Data and the Power of Gemini:

• Data is not limited to structured databases; it includes images, audio, and video.
• Google Cloud technologies natively support multimodal data, unlike "bolted-on" approaches.
• Access to data is governed by authentication and authorization.
• Having data is insufficient; it must be meaningful. Gemini, a cutting-edge LLM, provides thinking and reasoning capabilities.
• LLMs have evolved from being instruction-tuned to thinking and reasoning before providing answers.
• The ultimate goal is to automate repetitive tasks, enhancing productivity and creativity.

5. Compliance, Unified Search, and Agent Management:

• The solution supports compliance standards across various industries (e.g., ISO).
• It offers a single, unified search experience for external, internal, and private data.
• It serves as a central destination for managing agents created with different technologies (Salesforce, Oracle, Google Cloud, Microsoft OpenAI).
• Users can access and manage agents from a single dashboard, automating tasks.

6. NotebookLM and Language Support:

• NotebookLM, a previous Google tool, went viral due to its search capabilities and podcast creation feature.
• It creates human-like conversations, making research more engaging.
• The solution supports various industry tools (Microsoft Outlook, Teams) and is releasing connectors to different sources.
• It respects access control lists (ACLs) and integrates with existing security infrastructure (LDAP, Active Directory).
• It supports third-party identity providers (Microsoft Entra, Okta, Ping).

7. SAS Model and Data Indexing:

• The solution is offered as a SAS (Software as a Service) model, requiring a one-time effort to establish connections.
• Indexing is performed to rank pages, but it only filters content relevant to the specific user.
• It ensures that users only access information they are authorized to see, respecting data privacy.

8. Default Agents and Language Support:

• The platform provides access to the latest Google Cloud technologies, including Gemini versions.
• Two default agents are available: idea generation and research agent.
• The research agent can generate product design reports and audio versions.
• The solution supports major languages beyond English and Hindi.

9. Google's Infrastructure and Scalability:

• The solution leverages the same battle-tested cloud infrastructure that powers Google's consumer products (YouTube, Gmail, Search).
• This infrastructure supports over 2 billion users, ensuring scalability and reliability for agent space and generative AI applications.

10. Conclusion:

The speaker advocates for a solution that addresses data fragmentation, enhances productivity through agentic AI, and provides a secure, scalable, and user-friendly platform for enterprises to leverage the power of generative AI. The emphasis is on seamless integration, multimodal data support, and respecting existing security infrastructure. The platform aims to empower users to automate tasks, access relevant information, and make data-driven decisions.