Notes from production AI.
Long-form posts on RAG cost economics, MCP integration with enterprise systems, governance frameworks, and evaluation pipelines that survive contact with real data.
95 posts
The Small-World Revolution: Mapping the Seditious Metadata of 1775 Boston
A network science analysis of the Committees of Correspondence, Sons of Liberty, and Boston Masons to show how Paul Revere functioned as a high-betweenness-centrality node.
TabFM: Google's Zero-Shot Tabular Foundation Model for SPY Volatility
Google's TabFM turns tabular prediction into in-context learning. Here is what matters, why the non-commercial license matters, and how to test it on SPY volatility.
Black-Scholes Without the Magic: Turning Random Paths into a Pricing Equation
A story-driven derivation of Black-Scholes: how GBM, Itô calculus, delta hedging, and no-arbitrage turn a random stock path into an option pricing equation.
The Most Important Lemma of 2026 Was Written by an AI. There's No Way to Cite It.
GPT-5.5 suggested a four-line argument that helped disprove a 50-year-old conjecture. That argument is now un-citable. Here's what that means for the future of knowledge.
Geometric Brownian Motion in Python: The First Model Every Quant Learns
A practical, story-driven introduction to geometric Brownian motion: why quants model log returns, how volatility drag appears, and how to simulate price paths in Python.
From Itô to Black-Scholes: How Randomness Became a Pricing Engine
The bridge from stochastic calculus to quantitative finance: how Bachelier, Wiener, Itô, Black, Scholes, and Merton turned random price paths into a pricing engine.
Alignment Whack-a-Mole: Why Fine-Tuning Makes LLMs Regurgitate Your Copyrighted Books
Fine-tuning an LLM on a benign task (like expanding plot summaries) causes GPT-4o, Gemini 2.5, and DeepSeek-V3.1 to recall and reproduce up to 90% of copyrighted books from pretraining.
RAG Evaluation in Production: Moving Beyond Metrics for Enterprise Success
Your enterprise RAG app works in the lab, but how does it perform in production? Discover advanced strategies beyond static metrics to ensure reliability and business value.
From Wall Street to ERP: Why Transformer Embeddings Are Eating Enterprise Transaction Systems
Enterprise ERP logs are high-signal event streams. By applying sequence-based transformers (like Revolut’s PRAGMA) and Tabular Foundation Models (like Prior Labs’ TabPFN) to ERP data, we can build unified, multi-task systems that outperform custom feature engineering.
Why Your LLM is a Stochastic Process (And Why Temperature=0 Doesn't Save You)
A deep dive into why Large Language Models are auto-regressive Markov chains, how GPU floating-point non-associativity and FlashAttention break determinism, and why agent pipelines behave like controlled stochastic systems.
The ERP Copilot Security Dilemma: Dynamic Row-Level Security and Identity Delegation in LLMs
Naive Service Principal access exposes sensitive ERP data to LLM context windows. Learn how to architect zero-trust AI agents using OAuth2 On-Behalf-Of (OBO) token exchange and database-enforced Row-Level Security (RLS).
Beyond the Prompt: Context Engineering Patterns for Complex Enterprise APIs
Enterprise APIs are too massive for LLM prompts. Discover context engineering patterns like JIT schema pruning, semantic routing, and session compression to build efficient agents.
The Quiet Genius Who Made Randomness Calculable
Brownian motion had been observed, priced, explained, and made rigorous. Kiyosi Itô did something stranger: he built a calculus for paths too jagged to differentiate.
The One-Line Change That Beats Brute Force: Antithetic Variates in Monte Carlo Pricing
Monte Carlo option pricing has a brutal tax: to halve your error you need four times the paths. But one almost trivial idea — mirroring every random shock — buys up to 4.7× the accuracy for free. Meet antithetic variates.
The Unstoppable Mind: Norbert Wiener and the Mathematics of Continuous Randomness
A child who graduated university at 14, a PhD at 18, then years of wandering before MIT. Norbert Wiener didn't just tame randomness — he built the mathematical house it lives in, then invented cybernetics and warned the world about machines.
Dolt: What If Your Database Had a Git History?
Dolt is a MySQL-compatible database where every row has lineage, every schema change is a commit, and your entire data history is one SQL query away. Here is why 22,000+ developers are rethinking how they store data.
Microsoft Agent Framework 1.0 for .NET: The Agentic Runtime .NET Developers Have Been Waiting For
Microsoft Agent Framework 1.0 unifies AutoGen and Semantic Kernel into a single, production-ready agentic runtime for .NET. A deep-dive for product owners, architects, and developers — with a real expense-tracker agent built in C# to show how it all fits together.
Bachelier's Revenge: How a Forgotten Thesis Shaped Modern Finance
In 1900, a French orphan wrote a PhD thesis that predated Einstein by 5 years and seeded the Black-Scholes formula. Then the world forgot him. The story of how one equation — setting expectation to zero — became worth trillions.
D365 F&O Client & MCP Server v0.3.7: Request Tracing & Server Timing Are Here
D365 F&O Client & MCP Server v0.3.7 ships comprehensive request tracing and server timing visibility — making production debugging and performance analysis dramatically easier.
Let's Talk v0.2: Now Running on LangGraph 1.0+ 🚀
Let's Talk v0.2 is live at thedataguy.pro — upgraded to LangGraph 1.0+, LangChain 1.0+, Pandas 3.0, and a fully modernized stack. Here's what changed.
Mastering Observability in D365 F&O: A Complete Azure Application Insights Guide
Learn how to configure Azure Application Insights for Dynamics 365 F&O, write custom X++ telemetry, query with KQL, and build proactive alerting to achieve true operational observability.
When Randomness Met Calculus: A Practitioner's Guide to Stochastic Differential Equations
Itô gave randomness a mathematical language. Now let us use it. A hands-on guide to stochastic differential equations — what they mean, where they appear in physics, biology, and finance, and how to simulate them in Python with the Euler-Maruyama method.
From Pollen Grains to Nuclear Bombs: The Astonishing Story of Monte Carlo Methods
How a game of solitaire played by a sick mathematician in 1946 led to one of computing's most powerful techniques — and why randomness became the solution to the world's hardest problems.
From Symbolic Math to Random Sampling: Mastering Integral Calculations with Python
Explore three powerful approaches to solving integrals: analytical methods, Sympy symbolic integration, and Monte Carlo numerical methods. Learn how randomness can solve mathematical problems with stunning accuracy.
Claude’s Code Leaked: How to Build Your Own AI Coding Assistant from the Blueprint
Anthropic's "Claude Code" source was allegedly leaked. I'm breaking down its architecture and giving you a step-by-step guide to build your own AI software engineering tool, ethically and effectively.
Running Your Own D365 AI Assistant with Ollama: A Cost-Effectiveness Analysis
Discover when GPT-4o is overkill for D365 Finance & Operations. Learn how small, locally-hosted models running on Ollama can handle 70% of enterprise AI tasks at a fraction of the cost, with complete control over your data.
OWASP Top 10 for LLM Applications (2025): A Developer's Security Checklist
A practical security checklist for developers building LLM-powered applications. Covers all 10 OWASP risks for LLM apps (2025 edition) — prompt injection, data poisoning, excessive agency, Denial of Wallet, and more — with actionable mitigations you can implement today.
SymPy: Bridging the Math Gap in Gen AI Systems
Discover how SymPy transforms LLMs from approximate text predictors into precise mathematical reasoning engines. Learn to build neuro-symbolic AI systems that handle calculus, stochastic differential equations, and complex symbolic math with 100% accuracy.
Happy New Year 2026: Build 🏗️ Ship 🚢 Share 🚀
Kicking off 2026 with renewed energy and focus on what matters—building, shipping, and sharing. A New Year message to the AI and D365 community.
The Hidden Environmental Cost of AI: Who Really Pays for Your LLM?
Every LLM query has a carbon footprint. Discover the hidden environmental costs of AI systems, how they are externalized to the public, and practical strategies for building more sustainable enterprise AI solutions.
When in Doubt, Open It in Webtop: Browser Isolation for the AI Era
Discover how LinuxServer.io's Webtop 3.0 revolutionizes browser security and remote computing. Learn why isolated environments are critical in an age where AI-generated links and phishing attempts are increasingly sophisticated.
Building Production-Grade Video Search: HNSW, Vector Indexing, and Multimodal RAG
Master the infrastructure behind production video search systems. Learn HNSW graph indexing, hierarchical retrieval strategies, hybrid search, and VideoRAG architectures that power platforms like YouTube and TikTok at billion-video scale.
Video Embeddings at Scale: Mastering the Fourth Dimension with Spatiotemporal Transformers
Dive deep into the challenges of video embedding generation. Learn how VideoMAE V2s dual masking strategy and spatiotemporal transformers overcome the limitations of frame averaging to enable true temporal reasoning and action recognition.
Understanding Multimodal Embeddings: The Evolution from CLIP to Unified Foundation Models
Explore the paradigm shift in multimodal AI from isolated CLIP-style encoders to unified foundation models like Omni-Embed and VLM2Vec-V2. Learn how instruction-tuned transformers are revolutionizing cross-modal retrieval and embedding generation.
Node.js ORMs in 2025: Choosing Between Prisma, Drizzle, TypeORM, and Beyond
A comprehensive analysis of the Node.js ORM landscape in 2025. Compare Prisma, Drizzle, TypeORM, MikroORM, and Sequelize to make the right choice for your project.
OLMo 3: Charting a Path Through the Model Flow to Lead Open-Source AI
Discover OLMo 3 from Allen AI - a groundbreaking fully open language model family with complete transparency. Learn about model flow, available models, and how to leverage them in your projects.
d365fo-client v0.3.5: Intelligent Cross-Company Queries for D365 F&O
Automatic cross-company query detection for Dynamics 365 Finance & Operations - no configuration needed! Smart dataAreaId handling across Python API, CLI, and AI assistants.
Chronos-2: The Evolution from Univariate to Universal Time Series Forecasting
Amazon introduces Chronos-2, a foundation model that handles univariate, multivariate, and covariate-informed forecasting in a zero-shot manner using in-context learning.
AI-Powered Vendor Invoice Automation for D365 F&O: The First Step Toward Agentic AP
Introducing an AI prompt that extracts invoice data from PDFs and creates vendor invoices in D365 F&O automatically - the foundation for building autonomous AP agents with LangGraph.
D365 F&O MCP server v0.3.4 — SRS report generation, testing, and FastMCP improvements
d365fo-client v0.3.4 adds SRS (SQL Server Reporting Services) document generation tools, improved test coverage, FastMCP architecture updates, and better Azure AD integration for enterprise D365 F&O automation.
Building Intelligent D365FO Agents with LangGraph and Model Context Protocol
Learn how to create intelligent D365 Finance & Operations agents using LangGraph and Model Context Protocol. Mix custom tools with MCP tools, choose optimal models, and build automated ERP workflows.
Building an AI-Powered D365 F&O SSRS Report Download Agent with LangGraph
Learn how to build a conversational AI agent that downloads D365 F&O SSRS reports using LangGraph and local AI models. Streamlined implementation with 50% less code.
Streamline D365 F&O Data Migration with DMF: Complete Guide to Package Export and Import
Master the complete workflow for exporting and importing data packages between D365 Finance & Operations environments and legal entities using Data Management Framework (DMF) with automated sequencing, execution, and error handling.
One-Click Azure Deployment: D365FO MCP Server Gets Ridiculously Easy with API Key Authentication
Deploy D365FO MCP Server to Azure Container Apps in minutes with new automated scripts, ARM templates, and simplified API key authentication. No more deployment complexity!
D365FO MCP Server v0.3.0: Multi-Transport Support and OAuth 2.1 with Microsoft Entra ID
Discover D365FO MCP Server v0.3.0's game-changing features: multi-transport support (stdio/HTTP), OAuth 2.1 authentication with Microsoft Entra ID, enhanced Docker deployment, and comprehensive environment variable management.
How to Download Virtually Any SSRS Report from D365 Finance & Operations
Discover the hidden RunCopilotReport action that lets you programmatically download any SSRS document from D365 F&O - from invoices to purchase orders - with simple OData calls.
Complete Guide: D365 F&O MCP Authentication with User Credentials
Step-by-step guide to authenticate D365 Finance & Operations with Claude Desktop using Azure user credentials and MCP servers for AI-powered business intelligence.
D365FO MCP Server in Action: From Setup to Customer Intelligence in Minutes
Watch how D365 Finance & Operations MCP Server transforms complex enterprise data queries into natural conversations, demonstrated with real customer intelligence scenarios.
The Tool Problem: Why AI Assistants Are Only as Good as Their Tools
Claude Desktop can write brilliant code but can't analyze your customer data. Here's why comprehensive tool ecosystems are the secret to unlocking AI's true enterprise potential.
D365FO MCP Tools - From AI Assistant Dream to CFO Crisis Management
Join The Data Guy Show as Mo reveals how his custom-built MCP tools for AI assistants uncovered a $131K credit crisis in Microsoft's demo data, perfectly illustrating real-world enterprise scenarios.
D365FO-Client v0.2.3: Enterprise-Grade Credential Management and Advanced Sync Session Architecture
Deep dive into d365fo-client v0.2.3's revolutionary credential management system, advanced sync session architecture, and Azure Key Vault integration for enterprise D365 F&O operations.
Software 2.0 Meets Enterprise: Why AI is Eating Your ERP System
Explore how Andrej Karpathy's Software 2.0 paradigm is revolutionizing enterprise software, transforming complex ERP integrations into natural language conversations that democratize business intelligence.
Building the Future of D365 F&O Integration: AI-Powered Development with Model Context Protocol
Revolutionizing Dynamics 365 F&O integration with AI assistants: Introducing d365fo-client's Model Context Protocol server that enables natural language interactions with enterprise ERP systems.
Bridging Legal Requirements and Technical Implementation: A Practical Guide to AI Governance Frameworks
Explore how modern technology platforms like OpenMetadata, DataHub, and Apache Atlas enable organizations to meet evolving legal requirements from EU AI Act to NIST frameworks while building robust AI governance systems.
Open-Source Data Governance Frameworks: A Strategic Analysis of OpenMetadata, DataHub, Apache Atlas, and Amundsen
Comprehensive comparison of leading open-source data governance platforms - OpenMetadata, DataHub, Apache Atlas, and Amundsen - with practical guidance for strategic selection and implementation.
Let's Talk 0.1.5 Release: Enhanced Self-Hosting and Production-Ready Features
Announcing Let's Talk 0.1.5 with comprehensive self-hosting capabilities, enhanced security, modular architecture, and production-ready features for AI-powered interactive chat systems.
From Basic Tool Calling to Advanced ReAct Agents: A Complete Implementation Guide
Master the evolution from basic LLM tool calling to sophisticated ReAct agents. Learn hands-on implementation with OpenAI, LangChain, and LangGraph in this comprehensive tutorial.
Data Governance for AI and RAG Systems: A Strategic Imperative
Discover why specialized data governance frameworks are critical for AI and RAG systems, and learn practical strategies to ensure responsible, secure, and effective AI deployment.
Swagger vs. Scalar: A Comprehensive Analysis for API Documentation
A deep dive into Swagger and Scalar, comparing their features, performance, and ecosystem integration to help you choose the right API documentation tool.
Responsible RAG: Ethical Considerations in Retrieval-Augmented Generation
Explore the ethical landscape of Retrieval-Augmented Generation (RAG) systems—covering citation, attribution, bias, and transparency—and learn how to evaluate fairness and responsibility using Ragas metrics.
The Economics of RAG: Cost Optimization for Production Systems
A comprehensive guide to understanding and optimizing the costs of Retrieval-Augmented Generation (RAG) systems in production, from token usage and embedding storage to infrastructure and operational overhead.
Kimi K2: The Trillion-Parameter Open-Source Breakthrough That's Reshaping AI
Moonshot AI's Kimi K2 delivers trillion-parameter performance with groundbreaking agentic capabilities, challenging closed AI systems and democratizing frontier AI access.
Zero-Shot RAG Systems: The Data Guy Show Podcast Episode
Join Nazz and Mo on The Data Guy Show as they explore how to build Retrieval-Augmented Generation systems that work out-of-the-box with minimal tuning, featuring real-world examples and practical insights.
Context Engineering: The New Foundation for Scalable, Trustworthy AI
A comprehensive guide to Context Engineering, Learn why context is the backbone of modern AI, with references to deeper dives and podcasts.
Train Your AI Without Actually Training It: The Revolutionary Secret Every User Should Know
Discover the game-changing technique that turns any AI conversation into a personalized learning experience. No technical skills required - just smart prompting strategies that work.
Chain of Command in X++ vs Python: Extending Code the Safe Way
A deep dive into Microsoft D365 F&O's Chain of Command (CoC) in X++ and how it compares to Python's extensibility patterns. Learn the strengths, trade-offs, and best practices for safe, maintainable code extension.
26 Prompting Principles That Will Transform Your LLM Interactions
Discover the research-backed principles from 'Principled Instructions Are All You Need' that will revolutionize how you communicate with large language models like GPT, Claude, Gemini, and LLaMA.
What is Context in LLMs for End-Users? The Complete Guide
Understanding context in Large Language Models isn't just for developers. Learn what context really means for everyday AI users, why your ChatGPT 'forgets' things, and practical tips to get better results from any AI tool.
AI-Powered D365 F&O Code Review: Revolutionizing Enterprise Development Workflows
Discover how AI-powered code review tools are transforming Microsoft Dynamics 365 Finance & Operations development workflows with automated analysis, architecture compliance, and intelligent insights for enterprise applications.
You Can't Handle the Truth... Without Context!
Discover why context is the ultimate key to getting truthful, grounded answers from AI systems. Learn how proper context transforms LLM hallucinations into reliable, factual responses through real-world examples and practical techniques.
Introducing The Data Guy Show: Your New Favorite Podcast on AI, Data, and the Future
Announcing The Data Guy Show — a lively new podcast where Mo and Nazz break down the world of AI, data science, and agentic systems. Meet your hosts and discover what’s in store!
Best of both worlds: Comparing pnpm and uv Package Managers
A side-by-side comparison of pnpm and uv, two modern package managers for Node.js and Python, from installation to package management.
Zero-Shot RAG: Building Systems That Work Out-of-the-Box
Discover how to build Retrieval-Augmented Generation systems that perform effectively with minimal tuning, allowing for faster deployment and reduced development overhead while maintaining high quality responses.
Behind the Scenes of Let’s Talk: Building an AI-Powered Chat for Website Platform
A deep dive into the architecture, workflows, and technology powering Let’s Talk—an AI-driven Chat for Website platform designed for flexibility, transparency, and cost efficiency.
Announcing Let's Talk v0.1.1: Interactive AI Chat for Websites
Introducing Let's Talk v0.1.1 – a Svelte-based, open-source AI chat widget for websites, now live on TheDataGuy! Discover its features, integration, and how it powers interactive conversations.
It Depends on the Context: How LLMs Find the Truth
Explore why context is everything for large language models (LLMs), how it shapes their answers, and practical techniques to provide the right context for more truthful, reliable AI outputs.
Truth is Cold: LLM Temperature and Data-Driven Decision Making
Explore how the temperature parameter in large language models (LLMs) shapes the balance between truth, creativity, and data-driven decision making.
Build 🏗️ Ship 🚢 Share 🚀 Is the Way
How the 'Build, Ship, Share' mindset—championed by the AI Makerspace Bootcamp—reignited my passion for technology, learning, and community.
Evaluating Advanced RAG Retrievers: A Practical Comparison
A hands-on, metric-driven comparison of advanced retrieval strategies for RAG systems using LangChain and Ragas. See which retriever wins on accuracy, speed, and cost.
Building Your AI Data Moat: Competitive Advantage Through Proprietary Data
Learn concrete strategies for creating an unassailable competitive advantage through collecting, annotating, and protecting your organization's proprietary data assets in the AI era.
Introducing 'Let's Talk': An AI-Driven Chat Component for Technical Blogs
Discover how 'Let's Talk' transforms passive blog reading into interactive dialogue, enhancing information discovery and retention for technical content.
Metric-Driven Development: Make Smarter Decisions, Faster
Your Team's Secret Weapon for Cutting Through Noise and Driving Real Progress. Learn how to use clear metrics to eliminate guesswork and make faster, smarter progress in your projects.
Part 8: Building Feedback Loops with Ragas
A research-driven guide to designing robust, actionable feedback loops for LLM and RAG systems using Ragas. Learn how to select metrics, set baselines, define thresholds, and incorporate user and human feedback for continuous improvement.
Subscribe to Our Blog via RSS
Stay updated with our latest content by subscribing to our new RSS feed
Part 7: Integrations and Observability with Ragas
Discover how to generate robust test datasets for evaluating Retrieval-Augmented Generation systems using Ragas, including document-based, domain-specific, and adversarial test generation techniques.
Part 6: Evaluating AI Agents: Beyond Simple Answers with Ragas
Learn how to evaluate complex AI agents using Ragas' specialized metrics for goal accuracy, tool call accuracy, and topic adherence to build more reliable and effective agent-based applications.
Part 5: Advanced Metrics and Customization with Ragas
Explore advanced metrics and customization techniques in Ragas for evaluating LLM applications, including creating custom metrics, domain-specific evaluation, composite scoring, and best practices for building a comprehensive evaluation ecosystem.
Part 4: Generating Test Data with Ragas
Discover how to generate robust test datasets for evaluating Retrieval-Augmented Generation systems using Ragas, including document-based, domain-specific, and adversarial test generation techniques.
Part 3: Evaluating RAG Systems with Ragas
Learn specialized techniques for comprehensive evaluation of Retrieval-Augmented Generation systems using Ragas, including metrics for retrieval quality, generation quality, and end-to-end performance.
Part 2: Basic Evaluation Workflow with Ragas
Learn how to set up a basic evaluation workflow for LLM applications using Ragas. This guide walks you through data preparation, metric selection, and result analysis.
Part 1: Introduction to Ragas: The Essential Evaluation Framework for LLM Applications
Explore the essential evaluation framework for LLM applications with Ragas. Learn how to assess performance, ensure accuracy, and improve reliability in Retrieval-Augmented Generation systems.
Building Research Agent with RSS Feed Support
How I created a comprehensive research assistant that combines web search, academic papers, RSS feeds, and document analysis to revolutionize information discovery.
A C# Programmer's Perspective on LangChain Expression Language
My experiences transitioning from C# to LangChain Expression Language, exploring the pipe operator abstraction challenges and the surprising simplicity of parallel execution.
Data is King: Why Your Data Strategy IS Your Business Strategy
Discover why controlling unique, high-quality data is your organization's most valuable competitive advantage in the AI era, and how a strategic approach to data ownership is becoming essential to business success.
Coming Back to AI Roots - My Professional Journey
A personal reflection on my career journey from AI to web and enterprise software development, and why I'm returning to my original passion for artificial intelligence.