Mar '26 Regional Tech Events

User Groups

• Mar 3: Ohio North Database Training
• Mar 11: Azure Cleveland
• Mar 19: GLUG.NET
• Mar 25: Cleveland C# User Group

 

Conferences

• Feb – Jun: Agent Camp
• Mar 16: Memphis Agent Camp

 

Databricks Q&A

Yesterday I delivered a presentation on "Data Cleansing using Databricks".  Listed below are the questions that came up as well as the answers related to Databricks.

 

1. Does state transfer from one cell to another in the Databricks notebook? Yes, state does transfer from one cell to another, as long as you're running in the same notebook session and on the same cluster.

 

2. Can Databricks Jobs (aka "pipelines") be accessed through a notebook? Yes, you can fully interact with Jobs from a notebook using the Databricks REST API or the Databricks CLI.

Some of these interactions from a notebook include:

• Trigger a job run
• Check job run status
• Cancel a run
• Retrieve job metadata

 

However, you can't directly "open" the Jobs UI from a notebook, and you can't modify job definitions without using the API

 

Listed below is an example for accessing a Job from a notebook

import requests

import json

 

token = dbutils.secrets.get("my-scope", "my-token")

workspace = "https://<your-databricks-instance>"

 

job_id = 12345

 

resp = requests.post(

    f"{workspace}/api/2.1/jobs/run-now",

    headers={"Authorization": f"Bearer {token}"},

    json={"job_id": job_id}

)

 

print(resp.json())

 

3. Can one notebook call another notebook? Yes, you can use dbutils.notebook.run() to call other notebooks from Python code.

 

4. Can I import/export SQL Server data to/from Databricks database? Yes. This can be done using JDBC, Azure Data Factory, or Synapse to name a few options

 

5. How do I install and access the Databricks CLI?

From Windows Command Prompt:

> pip install databricks-cli

 

This will install databricks.exe in the same folder as the Python scripts

(i.e. C:\Users\JimSmith\AppData\Roaming\Python\Python313\Scripts).

If so, the following commands would be need to executed from that folder.

 

> databricks configure --token

 

You'll be prompted for:

Databricks workspace URL (Example: https://adb-123456789.12.azuredatabricks.net)

Personal Access Token.  This can be generated in Databricks by clicking

User Settings (User icon in upper right corner) → Developer → Generate New Token

 

After that, the CLI is ready to use.

> databricks clusters list

 

For more information, see What is the Databricks CLI?

Layers of RAG Architecture Patterns

Retrieval‑Augmented Generation (RAG) has become one of the most important design patterns in modern AI because it gives language models direct access to external knowledge.

Instead of relying solely on what a model has memorized during training, RAG systems retrieve relevant information from documents, databases, or other sources and feed it into the model at generation time.

This idea dramatically improves accuracy, reduces hallucinations, and allows AI systems to stay current without constant retraining.

 

RAG has evolved into a rich ecosystem of architectural layers, addressing different challenges:

1. Core Retrieval layer focuses on improving how information is found, from basic vector search to more advanced techniques like query expansion and hierarchical retrieval.
2. Structure‑Aware layer organizes and interprets data based on relationships, formats, or time, enabling retrieval from graphs, tables, or multimodal sources.
3. Reasoning‑Enhanced layer strengthens the model's ability to think with retrieved information through multi‑step search, agentic planning, self‑reflection, and high‑accuracy fusion.
4. System‑Level Orchestration layer coordinates multiple retrieval and reasoning strategies, integrating tools, memory, personalization, and routing to build adaptive, production‑ready AI systems.

 

Layer	Arch. Pattern	Pattern Description	Strengths	Weaknesses	Best Use
Core Retrieval	Basic RAG	Single-pass retrieval: embed query, retrieve top-k chunks, feed to LLM.	Simple, fast, easy to implement.	Weak with vague or ambiguous queries.	Baseline RAG, small datasets.
	Query Expansion RAG	Expands the user query into multiple variants to improve recall.	Handles vague or short queries well.	Can retrieve irrelevant results.	Search interfaces, consumer chatbots.
	Multi-Vector RAG	Stores multiple embeddings per document (sentence-level or attribute-level).	High precision for dense or multi-topic documents.	Higher storage and compute cost.	Technical manuals, scientific papers.
	Hybrid Search RAG	Combines vector search, keyword search, and metadata filters.	High recall and precision.	More complex retrieval logic.	Enterprise search, compliance.
	Cluster-Based RAG	Clusters documents and retrieves from the most relevant cluster.	Faster retrieval; scalable.	Cluster quality matters.	Large-scale corpora.
	Hierarchical RAG	Two-stage retrieval: coarse (document) then fine (paragraph or sentence).	Reduces noise, scales to long documents.	More complex pipeline.	Legal texts, long PDFs, structured corpora.
Structure Aware Retrieval	Graph-Based RAG	Converts data into a knowledge graph and retrieves via relationships.	Strong relational reasoning.	Requires graph construction and maintenance.	Enterprise knowledge bases.
	Chunk-Graph RAG	Builds a graph of chunk-to-chunk relationships for better navigation.	Strong for long or interconnected texts.	Requires preprocessing.	Books, manuals, long reports.
	Structured RAG	Retrieves structured data (tables, SQL, JSON) alongside text.	Accurate factual grounding.	Requires schema alignment.	Finance, logistics, analytics.
	Temporal RAG	Retrieval is time-aware (recentness, versioning, time decay).	Great for evolving data.	Requires timestamped corpora.	News, markets, real-time systems.
	Multimodal RAG	Retrieves images, audio, or video embeddings alongside text.	Richer context; cross-modal reasoning.	Requires multimodal indexing.	Vision-language agents.
Reasoning Enhanced Retrieval	Multi-Hop RAG	Performs sequential retrieval steps to answer multi-step questions.	Excellent for reasoning across documents.	Slower and more complex.	Research, academic QA.
	Agentic RAG	LLM plans retrieval steps and iteratively refines queries.	Strong for multi-step reasoning.	Expensive and harder to control.	Research workflows, complex tasks.
	Self-Reflective or Feedback-Loop RAG	LLM critiques its answer and triggers additional retrieval rounds.	Reduces hallucinations; improves reliability.	Higher latency.	High-stakes or regulated domains.
	Speculative RAG	LLM predicts what information it needs before retrieval.	Faster; reduces unnecessary retrieval.	Can mispredict needs.	Low-latency assistants.
	Fusion-in-Decoder RAG (FiD)	Encodes each retrieved chunk separately and fuses them during decoding.	Very high accuracy; handles many chunks.	Heavy compute cost.	High-quality QA systems.
	Retrieval-Graded RAG	Ranks retrieved chunks using a secondary scoring model or LLM.	Higher quality context.	Extra inference cost.	Precision-critical tasks.
System Level Orchestration	Routing or Mixture-of-Experts RAG	Router selects the best retriever or workflow for each query.	Domain-aware and flexible.	Requires router training.	Multi-domain assistants.
	Tool-Augmented RAG	LLM decides when to call external tools (SQL, APIs) alongside retrieval.	Strong for structured data.	Requires tool orchestration.	Analytics, BI, enterprise workflows.
	Memory-Augmented RAG	Stores long-term memory for retrieval (episodic or semantic).	Personalization and continuity.	Requires memory management.	Personal assistants, tutoring systems.
	Personalized RAG	Retrieval tuned to user profile or history.	Highly relevant results.	Requires user modeling.	Personalized assistants, education.
	Contextual RAG	Uses conversation history or metadata to refine retrieval.	Strong for multi-turn chat.	Can drift if context is noisy.	Customer support, assistants.
	Generative Index RAG	LLM generates synthetic summaries or embeddings to improve retrieval.	Better recall; compact indexes.	Risk of synthetic errors.	Large corpora with redundancy.

 

 

Linking one App.config in multiple projects

I recently had a .NET solution where I wanted to use the same App.config in 2 different projects.  I wanted to always ensure that changes made in the App.config of the original project would be reflected in other project(s) automatically.  Listed below are the steps I used to facilitate that process.

1.     Right-click your project in Solution Explorer

2.     Select "Add" -> "Existing Item..."

3.     Navigate to the file that you want to add to the solution

4.     [Important] Instead of hitting Enter or clicking the Add button, you want to click the down-arrow icon at the right edge of the Add button, and select "Add As Link".

 

 

Uploading documents to AI Foundry Agents

Q: Can we upload documents like pdf as input and configure the agents to retrieve the required content that is expected

 

A: The short answer is yes, but with some configuration.  Working in the Agent Playground, you have the option to “Add” Knowledge.  This knowledge can be from a variety of different data sources as seen below. As stated below “Currently only a single instance per each type of data source is supported.”  In my scenario, I had a single text file setup when I configured my agent, before being published.

 

Another option would be to utilize the Azure AI Search to index multiple documents from a data store, where documents can be uploaded after the agent is published.  As of today, there are 6 other options available for accessing documents

 

 

AI Agents vs. Agentic AI

AI agents and agentic AI are related but not the same. AI agents are task oriented systems built around LLMs, while agentic AI refers to a broader paradigm where AI systems exhibit autonomy, goal directed behavior, and self improving capabilities.

 

AI Agents vs. Agentic AI

AI Agents: perform tasks but do not necessarily set their own goals.

•             Modular systems built around LLMs or LIMs.

•             Designed for narrow, task specific automation.

•             Operate through tool integration, prompt engineering, and workflow orchestration.

 

Examples:

·                  A customer service chatbot

·                  A research assistant that retrieves and summarizes documents

·                  A coding agent that fixes bugs when prompted

 

Agentic AI: behave like agents (goal driven, adaptive, and self improving)

•             A broader paradigm where AI systems exhibit autonomy, self direction, and persistent goal pursuit.

•             Goes beyond task execution to include:

·                  Planning

·                  Reflection and self correction

·                  Long horizon reasoning

·                  Adaptive behavior

•             Often involves multi step, self initiated workflows.

 

 

Side by Side Comparison

 

Feature	AI Agents	Agentic AI
Scope	Narrow tasks	Broad, multi step goals
Autonomy	Low to moderate	High
Goal Setting	User defined	AI may refine or generate goals
Reasoning Depth	Shallow to moderate	Deep, reflective, iterative
Architecture	Modular workflows	Self directed cognitive loops
Examples	Chatbots, RPA-like tools	Autonomous research systems, self improving agents

 

Why the Distinction Matters

The research argues that the two concepts diverge in design philosophy and capabilities:

•             AI agents are an engineering pattern—a way to wrap LLMs in tools and workflows.

•             Agentic AI is a behavioral paradigm—systems that act with increasing independence.

 

This matters for:

•             Safety (agentic systems require stronger oversight)

•             Applications (agentic AI can handle long term, complex tasks)

•             Regulation (autonomy introduces new risks and responsibilities)

 

 

Examples to Make It Concrete

AI Agent Example

You ask: "Summarize these 10 PDFs."

The agent:

1. Retrieves files
2. Summarizes them
3. Returns results

It does not decide to read more papers or refine the topic unless instructed.

 

Agentic AI Example

You ask: "Research the best battery technology for drones."

An agentic system might:

1. Break the problem into sub goals
2. Search literature
3. Evaluate trade offs
4. Generate experiments
5. Identify missing data
6. Suggest next steps

It acts like a researcher, not just a tool.

 

Deployment Types in AI Foundry

Deploying a model in Azure AI Foundry can be done in 9 different ways.  Depending on the type of deployment chosen, it may impact one of more factors, such as cost, latency, efficiency for processing large datasets, compliance.  Listed below is a description of each deployment type, along with advantages and disadvantages.  For more details, please visit https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/deployment-types

Deployment Type	Description	Advantage	Disadvantage
Global Standard	Shared global infrastructure for general-purpose model inference.	Cost-effective and easy to scale.	Performance may vary under high demand.
Global Provisioned	Dedicated global infrastructure for consistent performance.	Reliable throughput and latency.	Higher cost due to dedicated resources.
Global Batch	Asynchronous global batch processing for large-scale inference jobs.	Efficient for processing large datasets.	Not suitable for real-time applications.
Data Zone Standard	Shared infrastructure within a specific data zone for compliance needs.	Meets data residency requirements affordably.	Limited performance consistency.
Data Zone Provisioned	Dedicated infrastructure in a data zone for high-performance workloads.	Combines compliance with consistent performance.	More expensive than shared options.
Data Zone Batch	Batch processing within a data zone for regulated data workflows.	Ideal for compliant, large-scale processing.	Slower response times; not real-time.
Standard	Default shared deployment for general use across Azure AI Foundry.	Simple setup and broad compatibility.	May lack advanced performance or compliance features.
Regional Provisioned	Dedicated infrastructure in a specific region for localized performance.	Optimized for regional latency and control.	Higher cost and limited to regional availability.
Developer (Fine-tuned)	Lightweight deployment for testing and iterating fine-tuned models.	Fast iteration and low cost for development.	Not suitable for production-scale workloads.

Nov '25 Regional Tech Events

User Groups

• Nov 4: Capital Area .NET Users Group
• Nov 12: Azure Cleveland
• Nov 13: Akron AITP
• Nov 19: Cleveland C# User Group
• Nov 20: GLUG.NET

 

Conferences

• Nov 11-13: .NET Conf 2025