Learning AI Evals with DeepEval, One Test at a Time
A practical introduction to evaluating AI answers with DeepEval, starting with an offline rule and progressing to a Codex-powered LLM judge.
Rustam Rakhimov
Software architecture, backend engineering, AI agents, and security
I am a software engineer who writes about software architecture, backend engineering, AI agents, and security. These are practical notes drawn from real work rather than tutorials. Views here are my own.
Featured writing
-
Architecture Is Becoming Part of the Implementation Process
· 6 min read · software-engineering architecture ai
Why AI-assisted prototyping makes architectural experimentation cheap enough to fold decision-making into implementation, and how ADRs still preserve the context.
-
Managing Agent State with ACID Principles
· 10 min read · ai software-engineering architecture
Applying database ACID principles - atomicity, consistency, isolation, and durability - to keep AI agent state recoverable across multi-step tasks.
-
Thinking Like a Modern Software Architect
· 3 min read · architect
Rising above the code to reason about systems, evolution, constraints, and trade-offs - the why, when, and to what extent of architectural decisions.
Latest posts
Building a Hello-World Custom Connector for Muse
Recently I wanted to discover how to build a custom connector that Muse — Meta’s personal AI agent — could use to reach a remote service of mine. The idea was simple: start with a temporary hello-world API as a proof of concept, prove the full round trip, and only then think about real functionality.
Architecture Is Becoming Part of the Implementation Process
Why AI-assisted prototyping makes architectural experimentation cheap enough to fold decision-making into implementation, and how ADRs still preserve the context.
Managing Agent State with ACID Principles
Applying database ACID principles - atomicity, consistency, isolation, and durability - to keep AI agent state recoverable across multi-step tasks.
My Experience with the Certified AI Security Professional (CAISP) Course
Artificial Intelligence is transforming every industry — but with innovation comes new security risks. Recently, I had the chance to explore these challenges firsthand through the Certified AI Security Professional (CAISP) course by Practical DevSecOps.
How the Gemini CLI Agent Thinks and Acts
Have you ever wondered what goes on inside an AI-powered command-line tool? How does it understand your requests, execute commands, and interact with your file system, all while feeling like a natural extension of your workflow?
Thinking Like a Modern Software Architect
Rising above the code to reason about systems, evolution, constraints, and trade-offs - the why, when, and to what extent of architectural decisions.
Building a Natural Language-Powered MCP SSE Server with Node.js
Recently, I built a custom MCP (Model Context Protocol) server using Node.js that supports Server-Sent Events (SSE) and an array of practical tools — including a powerful Natural Language to SQL (NL2SQL) converter. The final result was a lightweight yet extensible backend service that can be integrated with Claude Desktop, Windsurf AI, or any other MCP-compatible interface.
Testing Complex Logic Within Task Runners in Java
In this blog post, we’ll explore a common challenge faced by Java developers - testing complex logic within task runners. We’ll discuss a practical approach to isolate and test the logic within task runners effectively.
Rollout Logics Wrapper
In modern software development, the process extends beyond simply completing a task and moving on. It entails continuous work on new updates, features, and bug fixes. Therefore, the role of feature flags becomes crucial in the software’s lifecycle. Feature flags can be classified as either static or dynamic, depending on the speed at which their changes are applied to the application’s behavior. In this post, we will not discuss the various types of feature flags; instead, we will concentrate on the different approaches for writing code that depends on these variables and styles.