English
Deutsch
Polski
Serverless Microservices: Scaling on Google Cloud Run with GitHub Actions
A technical step-by-step tutorial on deploying and scaling serverless microservices on Google Cloud Run using GitHub Actions and Workload Identity Federation.
Cloud & AI
20 articles in this category
Follow this category by RSS
Building a RAG Knowledge Base with Google Cloud Vertex AI & BigQuery
A technical step-by-step tutorial on building a serverless RAG knowledge base using BigQuery Vector Search and Vertex AI Gemini for enterprise data.
Self-Hosted Large Language Model (LLM) with Ollama Behind an Auth Proxy
A technical step-by-step tutorial on deploying a private, GPU-accelerated LLM server using Ollama secured behind an Nginx authentication proxy.
Current AI Models Compared: GPT, Claude and Gemini versus DeepSeek and Qwen
Compare GPT-6, Claude 5.5, Gemini 3.8, DeepSeek and Qwen by access, tasks and local deployment.
Tutorial: Testing Chunk Size and Overlap Against a Retrieval Test Set
Chunk size is usually chosen once and never questioned, because questioning it seems to need a full evaluation. It needs thirty questions, one number, and a sweep that runs without a single call to a language model.
Chunk Size and Overlap in a RAG Knowledge Base: Effects on Embedding Count, Token Budget and Re-Indexing
Two numbers chosen before the first run decide how many embeddings exist, how much of every answer is context, and whether a changed setting means re-embedding everything. All three are arithmetic.
The Arithmetic Behind an LLM Bill
Input, output, cached input and the batch discount are four prices for the same request. Which one applies depends less on the model than on how the conversation is assembled.
GTM Integrations for Claude Code: Three MCP Servers, Their Tool Count and Their Context Cost
Changing a Tag Manager container from a prompt means handing a language model an OAuth token. Three MCP servers were started over stdio on 9 September 2026 and asked for their tool list: 18, 33 and 104 tools, and between 20 and 108 kilobytes of definitions that travel with every request. Underneath all three sits the same API with 106 methods, seven scopes and exactly two calls that reach the live container.
Tutorial: Measuring Token Consumption per Request and Enforcing a Monthly Budget
Every API response states exactly how many tokens it cost, in four separate classes. Logging that number instead of estimating it is twenty lines – and it is the only basis on which a budget can be more than a warning.
Bot Detection: IP Lists and Cryptographic Signatures Compared
Apple enlarged the Applebot address pool by roughly 194 per cent in August 2026. Every site verifying that crawler from a stored list had to fetch a new one. That maintenance burden is not an accident of this release; it is what address-based verification is.
