Zero PII Customer Intelligence - Part 2: Profiles, Signals & Segments

In Part 1 we covered the philosophy and series overview. In Part 1.1 we built the sample data generator. Now let's build the core system. This part focuses on: Zero-PII profile architecture - Ephemeral sessions vs persistent profiles Signals and weights - What…

"How do you work on all these projects and write these articles, is it just AI doing it all?"

Yes and no, I've ALWAYS been able to have multiple ideas in my head at the same time and let them coalesce until it's something I can build (I gave up all the 'social life' parts so it's not ALL GOOD!). I can write pages on each idea and DO all the time (check…

DocSummarizer Part 4 - Building RAG Pipelines

NuGet npm .NET Node.js This is Part 4 of the DocSummarizer series. See Part 1 for the architecture, Part 2 for the CLI tool, or Part 3 for the deep dive on embeddings. The hard part of RAG isn't the LLM. It's everything before the LLM. You've probably seen the…

Why You Probably Shouldn't Use Microservices (Yet)

Microservices have become the default "serious system" architecture. If you want to sound mature, you talk about service meshes, event buses, distributed tracing, and "independent deployability". If you want to look enterprise-ready, you draw boxes until the…

HTTP Over the Decades: A Story of Physics, Latency, and Grudging Adaptation

HTTP didn't evolve. It was forced to change by physics, latency, and misuse. Every version exists because the previous one hit a hard constraint. If you understand those constraints, you understand why the web works the way it does - and why so many "best…

No, Small Models Are Not the "Budget Option"

Small and local LLMs are often framed as the cheap alternative to frontier models. That framing is wrong. They are not a degraded version of the same thing. They are a different architectural choice, selected for control, predictability, and survivable failure…

GraphRAG Part 2: Minimum Viable GraphRAG

In Part 1, we explored why GraphRAG matters. Now let's build a minimum viable GraphRAG with three extraction modes: Mode LLM Calls Best For Heuristic (default) 0 per chunk Fast indexing, structured markdown Hybrid 1 per document Balance of speed and quality…

GraphRAG: Why Vector Search Breaks Down at the Corpus Level

Your RAG system is great at "needle" questions: retrieve a few relevant chunks and synthesise an answer. It struggles with two common query types: Sensemaking: "What are the main themes across this corpus?" Connective: "How does X relate to Y across different…

Probability Is Not a System: The Ten Commandments of LLM Use

I keep seeing the same failure mode in "AI-powered" systems: LLMs are being asked to do jobs we already solved decades ago - badly, probabilistically, and without guarantees. This isn't cutting edge. It's a regression. And this isn't a beginner mistake. The…

DataSummarizer: Fast Local Data Profiling

Most “chat with your data” systems make the same mistake: they treat an LLM as if it were a database. They shove rows into context, embed chunks, or pick “representative samples” and hope the model can infer structure, quality, and truth from anecdotes. It…

DocSummarizer Part 3 - Advanced Concepts: The "I Went Too Far 🤦" Deep Dive

This is Part 3 of the DocSummarizer series: Part 1: Building a Document Summarizer with RAG - The architecture and why the pipeline approach beats naive LLM calls Part 2: Using the Tool - Quick-start guide for the CLI Part 3: Advanced Concepts (this article) -…

DocSummarizer Part 2 - Using the Tool

GitHub release .NET Version This is Part 2 of the DocSummarizer series. See Part 1 for the architecture and patterns, or Part 3 for the deep technical dive into embeddings and retrieval. Turn documents or URLs into evidence-grounded summaries - for humans or…

Stop Shoving Documents Into LLMs: Build a Local Summarizer with Docling + RAG

Here's the mistake everyone makes with document summarization: they extract the text and send as much as fits to an LLM. The LLM does its best with whatever landed in context, structure gets flattened, and the summary gets increasingly generic as documents get…

Fetching and Analysing Web Content with LLMs in C#

📌 Note: This article teaches the fundamentals of web content extraction with LLMs using the simplest possible approach. For production use cases (web summarization, document analysis, agent tools), see DocSummarizer GitHub release - it implements the…

How to Analyse Large CSV Files with Local LLMs in C#

Series: Local LLMs for Data - Part 1 of 2 Here's the mistake everyone makes: they try to feed their CSV into an LLM. Don't. LLMs should generate queries, not consume data. You've got a 500MB CSV file and want to ask "What's the average order value by region?"…

Why I Don't Use LangChain (and What I Do Instead)

I'm a .NET developer. When I started building LLM-powered systems, everyone pointed me toward LangChain. "It's the standard," they said. "All the examples use it." And they were right - if you're in the Python ecosystem, LangChain is everywhere. But here's the…

Multiplatform AOT with SQLite: How to get it working!

Native AOT promises to turn your .NET applications into tiny, self-contained executables that start instantly and run anywhere—no runtime installation required. It's magic when it works. But add SQLite to the mix and suddenly you're debugging cryptic…

Building a "Lawyer GPT" for Your Blog - Part 9: Document Ingestion with Docling

Welcome to Part 9! In previous parts, we've built a robust RAG system that processes markdown blog posts and makes them searchable through semantic embeddings. Now it's time to expand our capabilities with Docling - a powerful document processing library that…

mostlylucid.ephemeral.complete; a strange concurrent systems pattern system in an LRU cache.

Well this has been my obsession for the past week. See the previous parts and what led to this; 'What if an LRU was an execution context.'. Now it's a set of 30 Nuget packages covering most major concurrent execution patterns (in TINY 5-10 line packages). Get…

Building a Frontend Before the API is Ready: Without Brittle Fixtures

How many times have you been blocked waiting for backend APIs to be ready? Or spent hours maintaining brittle mock data that becomes stale the moment requirements change? Enter mostlylucid.mockllmapi - a production-ready ASP.NET Core mocking platform that uses…