Articles & Feedback
Thoughts on backend development, software architecture, and AI integration in our workflows.
AI timeline
What has happened in artificial intelligence since 2017, and what each step actually changed. Every entry links to its primary source and carries the date it was checked.
An AI agent fleet in the wild: what the Tencent/Amap case teaches those who deploy agents
Agents running on Tencent Cloud got around Amap's protections using perfectly legitimate web tools. Egress filtering, public secrets, logging, attribution: a technical reading, without the hype.
We are generation zero
No one before us spent their days alongside software that writes code in their place. No hindsight, no established practice: opening a factual, sourced series on AI and its consequences.
1,567 zombie processes in a container
Inside a container, your application is process number 1, and number 1 has a chore it has no idea about. Sixteen days of accumulated garbage without a single metric moving.
Claude Sonnet 5: Anthropic Plays the Price Card Against Opus, GPT and Gemini
Performance close to Opus 4.8, five adjustable effort levels, and pricing half as expensive: a look at the benchmarks, pricing and implications for agentic AI projects.
Full LLM OCR POC: Vision Models Benchmark with Ollama
Complete benchmark of 4 vision models (llava, minicpm-v, qwen2.5vl, bakllava) for automatic data extraction from receipts and invoices. Architecture, results and Python implementation.
Complete RAG System: Architecture with FastAPI, HuggingFace, Ollama and Docker
Complete guide to building a production-ready RAG system: FastAPI, HuggingFace embeddings, Chroma DB, Ollama LLM and Docker. Applicable to any specialized document corpus.
Ollama as a local network service: deployment and integration
One GPU server, several clients over HTTP. Drivers, systemd, models, Aider integration, and the traps that make Ollama silently fall back to the CPU.
Prompting Best Practices with Claude Code
How to optimize your Claude Code usage: context management, disabling unused MCPs, and effective prompting patterns to cut your token consumption by 2x or 3x.
Playwright and Chromium in Docker Production: A Practical Guide
Robust Playwright and Chromium configuration in Docker: Dockerfile, non-root permissions, memory management and debugging common errors.