Skip to content
imarch.dev
Ilyas Mustafin

Ilyas Mustafin

CTO / Enterprise Architect

Notes on architecture, DevOps and building products

Topics
New FinOps in Practice. How to Stop Burning Money in the Cloud

FinOps in Practice. How to Stop Burning Money in the Cloud

A practical guide to cloud cost optimization: 6 patterns, free tools (Infracost, OpenCost, cloud-nuke), a one-day checklist, and real savings numbers.

DevOps cloud FinOps
Your AI Assistant Gets Dumber Every Minute

Your AI Assistant Gets Dumber Every Minute

Breaking down GSD - a system for fighting context degradation when working with AI. Spec-driven approach, fresh contexts and atomic commits instead of chat chaos.

AI development tools
Agentic Reality Check: Only 11% of Companies Use AI Agents in Production

Agentic Reality Check: Only 11% of Companies Use AI Agents in Production

Breaking down the Deloitte Tech Trends 2026 report on AI agents. Real adoption numbers, Toyota, HPE, Dell, and Moderna case studies, and why most agentic projects are doomed to fail.

AI architecture transformation
AI Inference and Cloud-Native: Kubernetes as AI Hub

AI Inference and Cloud-Native: Kubernetes as AI Hub

The CNCF executive director predicts massive growth in cloud-native software consumption driven by AI inference. What this means for architects and platform engineering right now.

platform-engineering kubernetes ai-inference
Amazon and OpenAI: $50B and AI Agent Architecture

Amazon and OpenAI: $50B and AI Agent Architecture

Amazon invests $50B in OpenAI and becomes the exclusive cloud provider for Frontier. What does this mean for architects building enterprise AI right now?

aws openai cloud
qwen3 vs nomic: swapping embedding models with real numbers

qwen3 vs nomic: swapping embedding models with real numbers

nomic-embed-text was terrible at finding Russian content - the target chunk ranked 44th out of 48. Switched to qwen3-embedding - same chunk jumped to #1. Full benchmark on live chatbot data.

ai RAG architecture
BPM Platforms: Evolution or Archaism?

BPM Platforms: Evolution or Archaism?

800 IT leaders from large enterprises told Camunda about the real state of process automation. The numbers are thought-provoking - and not just about BPM.

architecture BPM transformation
Hybrid RAG: How the Bot Lost 80% of Its Weight

Hybrid RAG: How the Bot Lost 80% of Its Weight

The chatbot was sending 6,000 tokens per request - all 26 articles, entire career, all services. Hybrid RAG kept a stable core and taught the bot to recall only what's relevant. Real metrics, architecture, and code.

ai RAG architecture
Four Bugs in One Evening

Four Bugs in One Evening

The bot lost sessions, switched languages, hallucinated email addresses, and sent messages to spam. Four bugs, four root causes, four fixes.

ai debugging claude code