IBM MMA - Power efficiency...

Equitus's KoGen (knowledge generation) - KGNN (Knowledge Graph Neural Network) can enhance IBM Power11 users by leveraging the system's specialized hardware and software to create a highly efficient and intelligent AI ecosystem.

This combination allows for on-premises AI that is powerful, secure, and integrated with existing enterprise data, without the need for additional GPUs.

---

How KGNN Uses IBM Power11 Hardware

Equitus's KGNN is specifically optimized to take advantage of IBM Power11's unique architecture. Instead of relying on expensive, power-hungry GPUs, KGNN uses the Matrix Math Accelerator (MMA) built directly into the Power11 processor.
This allows for:

• High-Performance AI Inference: The MMA accelerates matrix multiplication, a core operation in deep learning and AI. KGNN leverages this to perform real-time AI inferencing directly on the CPU, which is crucial for applications that require low latency. 🚀

• Reduced Cost and Energy: By eliminating the need for separate GPUs, organizations can significantly lower their hardware and operational costs.⁴This also makes the solution more energy-efficient, supporting sustainable IT initiatives.

• On-Premises Security: Running AI workloads on-premise keeps sensitive data within the organization's control, addressing data sovereignty and privacy concerns.⁵This is a critical advantage for regulated industries and government agencies.

---

Function Calling and A2A Protocols with KGNN

Equitus's KGNN can work with function calling and Agent-to-Agent (A2A) protocols to create a sophisticated, multi-agent AI system on IBM Power11.

• Function Calling: In this model, an AI agent, such as a Large Language Model (LLM), uses function calling to interact with external tools.⁶KGNN can serve as a powerful tool or knowledge base that the LLM calls upon. For example, an LLM could use function calling to query the KGNN for a specific piece of information. The KGNN would then use its structured knowledge graph to provide a precise, factual answer, effectively acting as an intelligent RAG (Retrieval-Augmented Generation) engine.⁷This prevents hallucinations and grounds the LLM's responses in a permanent, verifiable memory.

• Agent-to-Agent (A2A) Protocol: A2A protocols enable multiple, specialized AI agents to collaborate on complex tasks.⁸KGNN can act as a central knowledge hub that different agents can access and contribute to. For instance, a "data ingestion agent" could use the A2A protocol to add new information to the KGNN, while a "report generation agent" could then query the KGNN to produce a detailed report. This allows for a modular, scalable AI workflow where each agent is an expert in its domain, and the KGNN provides the shared context and memory for their collaborative work.

---

The Role of MCP (Model Context Protocol)

While often used in a similar context to function calling, MCP is a standardized protocol for connecting AI models to external tools.⁹KGNN can be an ideal tool within an MCP framework, as it provides a structured, context-rich environment that any MCP-compliant model can easily query. MCP allows for a unified interface, meaning that once KGNN is integrated into the protocol, any model that understands MCP can access its knowledge graph without custom coding.¹⁰This promotes interoperability and scalability across a diverse AI ecosystem on IBM Power11.