Rob Taylor • .NET Developer • Entrepreneur • AI Cloud Enthusiast
07-05-2026
Welcome! I’m a software engineer with over 20 years of experience in the Microsoft
ecosystem. This blog chronicles my journey as I navigate the shifting world of AI,
diving deeper into modern cloud and AI implementations. I will also track major news
and events to bring you the latest developments in the AI race and how they affect you.
Follow along for
practical insights on configuring Azure clouds and building
responsible AI solutions with RAG pipelines and pre-built LLMs.
Check out the About page
to learn more about my background.
In a surprising move, Microsoft CEO Satya Nadella shared an essay with the IT community last weekend that echoed the sentiments of previous AI leaders, such as Palantir's Alex Karp. While his essay left many on social media questioning his true motives, one thing is certain: his warnings about data risks are absolutely true.
Kicking off a new project with a new mindset: Multicloud AI. In this post, I cover step 1 (completed) to find a solid solution to database integration and interoperability with a LLM and use it as an opportunity to pivot toward the future of what AI cloud architecture may look like soon.
Probably the worst programming technology I have ever used is the SQL MCP container. I don't even know why it exists. If you think it will mimic the power of an LLM or a SQL database, it won't. I will forge on and mix things up to find a better solution, but I hope to save someone else the trouble.
Connecting the SQL Model Context Protocol (MCP) came with a few challenges, and at the end of the day, SQL MCP was not for me. However, that doesn't mean it's useless for everyone. If you are encountering errors with your container, it may be the dab-config.json file. This guide can help you fix it.
Durable Functions became a handy method for running scheduled, orchestrated, in-context RAG requests to generate content, but I was not able to use one of its most beneficial features: parallel execution. Still, I didn't walk away disappointed, and I have several optimization suggestions to share.
Copilot has been a great friend at those pertinent times when a boost to productivity is in order. Boilerplate creation on a new project can save many hours, and its ability to guide voyages into the unknown has significantly sped up my learning process. However, as good as Copilot is, you have to be even better when reviewing its code.
I, like many others, am digging deep to get a true bearing on generative AI's capabilities, accuracy, cost, and integration processes. You do wonder what is going on out there - what is being planned, and who is planning it. Palantir CEO Alex Karp gave us his assessment the other day on CNBC, and it sure as hell isn't all peaches and roses.
I posted a blog the other day about my in-context RAG project, which is finally completed. After using it, I realized a few things regarding it's effectiveness, when it should be used and how specific platform settings can drastically alter your processing time
Initial thoughts on leveraging the Code Interpreter tool for highly accurate data validation and statistical calculations: alerting users to inline data anomalies is straightforward, but orchestrating automated, on-the-fly text corrections requires a bit more architectural complexity.
I am setting up an In-Context RAG pipeline that utilizes Web Search (formerly known as "Grounded with Bing"), pre-loading it with my own data as opposed to using public-facing prompts. Quite honestly, it has been an exciting experience, and I wanted to pass along some important information regarding agent settings for those looking to do something similar.