VB
Victor Buzin

All Posts

Thoughts on software engineering, architecture, and the .NET ecosystem

RAG Evaluation Before Production: A Practical Guide
2026-08-0415 min read

RAG Evaluation Before Production: A Practical Guide

A RAG system without a maintained evaluation dataset is a system that cannot be improved or safely changed. A practical guide to the Minimum Viable Evaluation Dataset (MVED), the five essential categories, LLM-as-Judge architectures, and connecting evaluation to deployment gates.

AIArchitectureLLM
Read Article
Chunking Is Not a Strategy: How to Choose Retrieval Units
2026-08-0110 min read

Chunking Is Not a Strategy: How to Choose Retrieval Units

The choice of retrieval unit is the most consequential design decision in a RAG pipeline. Why character splitting fails, the precision/recall trade-off, parent-child and hierarchical patterns, the lost-in-the-middle constraint, and a unit-selection table.

AIArchitectureLLM
Read Article
From Senior Engineer to Principal: What Changes and What Does Not
2026-07-2912 min read

From Senior Engineer to Principal: What Changes and What Does Not

The jump from Senior to Principal is not a linear climb — it's a career pivot. A field note on impact radius, problem sourcing, ambiguity, and the day-to-day reality of moving from individual execution to organizational leverage.

CareerLeadershipEngineering Management
Read Article
RAG Is a System, Not a Prompt: Retrieval, Context, Evaluation and Operations
2026-07-2622 min read

RAG Is a System, Not a Prompt: Retrieval, Context, Evaluation and Operations

RAG answer quality is determined at four separate stages — Retrieval, Context, Answer, Operations — each testable independently. A stage-by-stage framework with metrics, ten common failure modes, and a senior-engineer checklist.

AIArchitectureLLM
Read Article
5 Questions Before Starting a RAG PoC
2026-07-238 min read

5 Questions Before Starting a RAG PoC

Most RAG PoCs fail not because the technology is broken, but because the architecture was built on unexamined assumptions. Five questions every technical lead should answer before writing the first line of orchestration code — covering retrieval units, evaluation, corpus ownership, latency/cost, and failure UX.

AIArchitectureLLM
Read Article
A Practical Reference Architecture for Enterprise RAG on .NET and Azure
2026-07-2015 min read

A Practical Reference Architecture for Enterprise RAG on .NET and Azure

Enterprise RAG on .NET is a system with eight distinct decision surfaces, each with its own failure modes. A staff-level reference architecture covering ingestion, vector stores, hybrid retrieval, reranking, and observability on Azure.

AIArchitectureLLM
Read Article
From AI Demo to Production: What Enterprise Teams Actually Need
2026-07-179 min read

From AI Demo to Production: What Enterprise Teams Actually Need

A field guide for CTOs on closing the gap between a working AI demo and a governed, production-ready enterprise system — covering architecture, reliability, security, cost, and ownership.

AIArchitectureLLM
Read Article
Why 'an LLM Chatbot' Is Not an AI Architecture
2026-07-177 min read

Why 'an LLM Chatbot' Is Not an AI Architecture

An LLM call is not an architecture. A field guide for technical leads and architects on the seven non-negotiable surfaces — retrieval, evaluation, economics, ownership, and failure UX — that separate a prototype from a production system.

AIArchitectureLLM
Read Article
.NET 11: A Deep Dive into the Future of .NET
2026-04-2512 min read

.NET 11: A Deep Dive into the Future of .NET

Explore the exciting new features in .NET 11 — from runtime-native async and C# 15 to massive performance improvements and ASP.NET Core 11 enhancements.

.NETC#ASP.NET Core
Read Article
RSS Feed