🇷🇺 Russian Version

Engineer-practitioner. Civil engineering technician (PGS) by education, systems researcher by calling.

I build software systems with the same level of responsibility I would apply to construction: focus on structure, protocols, and long-term reliability[cite: 1]. My path is one of "learning-by-doing," supported by rigorous engineering protocols and living documentation.

• MSCodeBase — Extending Zed IDE. My main "proving ground" where I implement AST-aware search, LanceDB, and complex indexing pipelines.
• gemma_agent — AI agent experiments. Exploring where LLM "magic" ends and strict protocol constraints begin.

I don't just write code — I build environments that adhere to strict principles:

• Clean Architecture (The "Tumbler" pattern: technology shouldn't dictate architecture).
• Transparency (Every architectural decision is backed by experiments recorded in EXPERIMENTS_LOG.md).
• Safety (Code without tests is incomplete construction)[cite: 1].
• Living Documentation (Code and documentation are synchronized in every commit).

Detailed breakdowns of specific technical discoveries and architectural challenges encountered during the development of this project:

• 📊 PageRank vs RAG on a Real Codebase: Corrected Numbers, and What I Almost Got Wrong Twice
  A head-to-head comparison between semantic (Vector RAG) and structural codebase retrieval methods, featuring two rounds of honest metric self-correction.

• 🛡️ I Asked One AI to Fact-Check Another AI's Audit of My Own Code
  A real-world case study on using cross-verification to defend against high-confidence LLM hallucinations during security code reviews.

• ⚡ The Silent Vector Contamination Bug: Why Your Concurrent Embeddings Might Be Lying to You
  A deep dive into thread-safety and memory isolation issues when generating concurrent embeddings locally using Intel OpenVINO in Python.

📊 GitHub Stats