Posts

5 Advantages of Granite 4.1 LLMs

-
Granite 4.1 is IBM’s new family of dense decoder ‑ only LLMs (3B, 8B, 30B) trained on ~15 trillion tokens with a five ‑ phase pre ‑ training pipeline, followed by 4.1M curated SFT (Supervised Fine Tuning). The family of models is released under Apache 2.0.   Granite   4.1 models consistently match or outperform larger competitors, with lower hardware requirements: 30B model outperforms Google’s Gemma ‑ 4 ‑ 31B ‑ it 8B model beats Gemma ‑ 4 ‑ 26B ‑ A4B ‑ it Dense architecture ensures predictable latency and stable token usage   Enterprise ‑ Grade Predictable Inference Granite   4.1 is designed for real ‑ world business workloads where speed, cost, and determinism matter. Strong instruction ‑ following and tool ‑ calling without long chains of thought. Dense models avoid the variability of MoE (Mixture-of-Experts) routing FP8 quantization options reduce memory footprint while preserving accuracy.   High ‑...
Post a Comment
Read more

AIProjectClient vs. Azure Open AI client

-
Image
Overview AIProjectClient is the unified Azure AI Foundry project SDK. It's for full Azure AI Foundry project management + agents + datasets + indexes + evaluations + OpenAI client generation. Azure OpenAI Client for Direct model inference (chat, embeddings, images) using OpenAI ‑ compatible endpoints.   Capabilities AIProjectClient provides control in the following areas: 1. Full lifecycle Agents: Create, update, delete, and run Azure AI Agents.  Azure OpenAI client cannot do this.  2. Datasets & File Management: Upload documents, create datasets, and use them in agents or evaluations.  3. Search Indexes: Create and manage RAG indexes inside your project.  4. Evaluations: Rules, taxonomies, evaluators, and insights for model quality.  5. Unified OpenAI Client: AIProjectClient can generate a fully configured OpenAI client (`get_openai_client()`), so no need to manage separate credentials....
Post a Comment
Read more

MS Foundry Developer Migration Checklist

-
Image
Microsoft Foundry recently implemented a variety of changes across various areas. To assist with these changes, a Developer Migration Checklist is provided below to help maneuver these changes.   1. Migrate to the Unified SDK 2.0 [] Replace all uses of the old azure-ai-agents package with azure-ai-projects 2.0. [] Update code to use AIProjectClient for model inference, agents, evaluations, memory, and tracing. [] Remove legacy preview flags and update any custom tool or MCP integrations. [] Validate that all tool schemas and agent configurations work under the new client.   2. Move Agents to the Foundry Agent Service (GA) [] Migrate existing agent deployments to the new Agent Service runtime. [] Update agent code to use the OpenAI Responses compatible interface. [] Reconfigure private networking, Entra RBAC, and tracing endpoints. [] Test agent behavior in the updated Agent Playground and tracing...
Post a Comment
Read more

What is Query Boosting, Weighting, and Thresholding?

-
Image
Query Boosting means increasing the importance of certain terms or fields in a search query so they influence the ranking more strongly. Sometimes not all parts of a query are equally important. For example: - In a product search, matching the title might matter more than matching the description. - In a document search, matching a keyword might matter more than matching the body text.   For example, if you search for: title:"machine learning"^3 description:"machine learning" The "^3" means “boost the title match 3× more than the description match.”     Weighting is the general idea of assigning different levels of importance to features, fields, or signals during ranking or scoring. Boosting is a type of weighting, but weighting can apply to: - Query terms  - Document fields  - Machine ‑ learning features  - User behavior signals (clicks, recency, popul...
Post a Comment
Read more

All About AI Foundry Agents

-
Image
In a recent episode of "Made for Dev" with Stephen Simon, I get into the world of AI Agents in Microsoft Foundry. I explain why agents are the "new microservices" and demonstrate how to build, configure, and orchestrate them using natural language.   What you’ll learn in this video: • What is an AI Agent? Understand the shift from LLMs that just answer questions to agents that execute business tasks [ 01:30 ]. • The "Physician" Analogy: Why an LLM is like a doctor who knows the theory but needs "tools" to actually take your blood pressure [ 02:05 ]. • Step-by-Step Demo: Watch Sam build a weather-forecasting agent from scratch in Azure AI Foundry, including tool integration via Logic Apps [ 04:54 ]. • Evaluation & Quality: How to interpret "Task Adherence" and "Intent Resolution" scores to ensure your agent is performing accurately [ 10:34 ]. • Advanced Orchestra...
Post a Comment
Read more

Copilot Studio vs. Microsoft Foundry

-
Image
One of the questions many tech leaders ask is “Which tool should I use for AI applications, Copilot Studio or Microsoft Foundry?”   Copilot Studio is a Low ‑ code, business ‑ friendly tool for the not-very technical user.  It’s fast to deploy and accessible at https://copilotstudio.microsoft.com   Microsoft Foundry (formerly known as Azure AI Foundry) is a Pro ‑ code tool designed to be more developer ‑ centric for the more technical user.  It offers full control, scalability, and is Azure native but accessed from a different URL ( https://ai.azure.com )   Summary    
Post a Comment
Read more

Handling Missing Data - Part 1

-
One of the big topics I have been speaking about is “ Data Cleansing for Machine Learning ”.  A major part of the data cleansing process is filling in missing values.  These values could be missing for a variety of reasons (manually entered data, improper data collection, incorrect exception handling, etc.)  Missing data can be summed up into 3 categories: MAR: Missing at Random, for patterns tied to known variables (e.g., age, education) where someone did not want to answer a question. MCAR: Missing Completely at Random, for random glitches, accidental loss (i.e. sensor failure) MNAR: Missing Not at Random, describes a situation where data is missing because of the value of the missing data itself.   To counter the issue of missing data, several techniques are available to fill in missing data.  In my video , I discuss 3 of those techniques. Multiple Imputation by Chained Equations (MICE) handles numeric or...
Post a Comment
Read more