I want to share my experience in the cloud, AI, and agents development areas.
The post explains why traditional Retrieval-Augmented Generation (RAG) approaches no longer scale and how modern architectures, including GraphRAG, address these limitations. It highlights why data quality, metadata, and disciplined system design matter more than models or frameworks, and provides a practical foundation for building robust RAG systems, illustrated with IBM technologies but applicable far beyond them.
This post is about setup to utilize Granite 4 models in Ollama for VS Code with watsonx Code Assistant. The process includes inspecting available models, uninstalling old versions, installing new models, and configuring them for effective use. The experience emphasizes exploration and learning in a private, efficient AI development environment.
The Model Context Protocol (MCP) is being increasingly utilized in AI applications, particularly with the watsonx Orchestrate ADK. This setup allows users to develop and manage agents and tools through a seamless integration of the MCP server and the Development Edition, enhancing user interaction and functionality in coding environments.
This blog post outlines a practical example of setting up a custom component in Langflow to connect with an external weather API and import it into the watsonx Orchestrate Development Edition. The process emphasizes learning through experimentation rather than achieving a flawless solution, highlighting the potential of Langflow and watsonx Orchestrate for AI development.
This post explores how Generative AI and Agentic AI emphasize trustworthiness over absolute determinism. As AI's role in enterprises evolves, organizations must focus on building reliable systems that operate under risk, balancing innovation with accountability. A personal perspective.
The post discusses integrating Knowledge Graphs with Retrieval-Augmented Generation (RAG), specifically using Neo4j and LangGraph. It outlines an example setup where extracted document data forms a structured graph for querying. The system enables natural question-and-answer interactions through AI, enhancing information retrieval with graph relationships and embeddings.
This post shows how to use Langflow with watsonx.ai and a custom component for a “Temperature Service” that fetches and ranks live city temperatures. It covers installation, flow setup, agent prompting, tool integration, and interactive testing. Langflow’s visual design, MCP support, and extensibility offer rapid prototyping; future focus includes DevOps and version control.
The post explores the integration of Large Language Models (LLMs) in applications, stressing the need for effective production testing. It introduces the AI Operational Complexity Cube concept, emphasizing new testing dimensions for LLMs, including prompt testing and user engagement. A structured testing approach is proposed to ensure reliability and robustness.
This blog post explains how to deploy a fine-tuned model to IBM watsonx on IBM Cloud. It highlights the advantages of using this platform, such as avoiding infrastructure management and ensuring enterprise security, as well as detailed steps for configuration, deployment, and accessing the model from IBM watsonx.
The blog post outlines updates on fine-tuning a model with the InstructLab , detailing tasks like data preparation, validation, synthetic data generation, model training, and testing. It emphasizes the need for extensive and accurate input for effective training, while only minimal changes in the overall process since previous versions, particularly in handling data quality. This blog post contains updates related to my blog post InstructLab and Taxonomy tree: LLM Foundation Model Fine-tuning Guide | Musician Example.
Thomas Suedbroecker's Blog 