# Vitalii Cherepanov

> 15+ years senior backend engineer specialising in AI-native tooling, distributed systems, and developer productivity. Building infrastructure for the AI coding agent era.

Personal site of Vitalii Cherepanov — Principal Backend Engineer · AI Infrastructure Architect · Open Source Builder.
Based in Železnička 20, Novi Sad, Serbia.
Email: vitalii@vbcherepanov.com.

This is the backend / AI infrastructure Vitalii Cherepanov (`vbcherepanov` on GitHub, Medium, and Dev.to). Not to be confused with academic publications by a different Vitalii Cherepanov, a Candidate of Economic Sciences who writes about digital transformation of industry — that is a separate person.

## About

- [Home (English)](https://vbcherepanov.com/)
- [Home (Русский)](https://vbcherepanov.com/ru)
- [Home (Srpski)](https://vbcherepanov.com/sr)

## Open source projects

- AISWARM (in progress (testing)): Distributed cognitive workforce runtime — the AISWARM + a2abridge + BrainCore triad.
  - Case study: https://vbcherepanov.com/projects/aiswarm
- [total-agent-memory](https://github.com/vbcherepanov/total-agent-memory): Persistent memory for Claude Code & Codex CLI.
  - Case study: https://vbcherepanov.com/projects/total-agent-memory
  - Homepage: https://totalmemory.dev
- [braincore](https://github.com/vbcherepanov/braincore): Local-first cognitive memory for AI coding agents.
  - Case study: https://vbcherepanov.com/projects/braincore
  - Homepage: https://getbraincore.com
- [claude-statusbar](https://github.com/vbcherepanov/claude-statusbar): Rich two-line status bar for Claude Code CLI.
  - Case study: https://vbcherepanov.com/projects/claude-statusbar
- [a2abridge](https://github.com/vbcherepanov/a2abridge): Open A2A 1.0 mesh for Claude Code, Codex, Cursor, Cline, Continue, Gemini.
  - Case study: https://vbcherepanov.com/projects/a2abridge
- [phpfilebroker](https://github.com/vbcherepanov/phpfilebroker): Filesystem as a message broker — exchanges, streams, dead letters.
  - Case study: https://vbcherepanov.com/projects/phpfilebroker
- [b-docker](https://github.com/vbcherepanov/b-docker): Bitrix Docker tooling that finally feels modern.
  - Case study: https://vbcherepanov.com/projects/b-docker
- [php-llamacpp-benchmarks](https://github.com/vbcherepanov/php-llamacpp-benchmarks): Six llama.cpp-inspired optimisation patterns, measured in PHP 8.4.
  - Case study: https://vbcherepanov.com/projects/php-llamacpp-benchmarks

## Services

Independent contractor. Email orders to vitalii@vbcherepanov.com.
- [All services](https://vbcherepanov.com/services)
- Websites & landing pages: Fast, multilingual, SEO-ready — Lighthouse 100/100/100/100.
  - Details: https://vbcherepanov.com/services#websites
- Backend services & APIs: Go and PHP/Symfony backends that survive production.
  - Details: https://vbcherepanov.com/services#backend
- AI integration: Plug LLMs into your product — without the demo-grade brittleness.
  - Details: https://vbcherepanov.com/services#ai-integration
- RAG systems: Retrieval that actually retrieves — same recipe as total-agent-memory (R@5 = 97 %).
  - Details: https://vbcherepanov.com/services#rag
- Clean up AI-generated codebases: Refactor vibe-coded sprawl into something you can ship and maintain.
  - Details: https://vbcherepanov.com/services#ai-cleanup
- Tune AI agents to write production-grade code: Claude Code / Cursor / Codex / Cline — wired with memory, hooks, MCP and a feedback loop that catches the bad output before you do.
  - Details: https://vbcherepanov.com/services#agent-tuning
- Architecture & strategy consulting: Second opinion on stack, architecture, AI strategy — by the hour or per engagement.
  - Details: https://vbcherepanov.com/services#consulting

## Articles

- [I Scaled PHP Until It Broke. Three llama.cpp Patterns Saved It.](https://vbcherepanov.com/articles/i-scaled-php-until-it-broke)
  - Canonical (medium): https://medium.com/@vbcherepanov/i-scaled-php-until-it-broke-three-llama-cpp-patterns-saved-it-12ddb096ab32
  - Six llama.cpp optimisations translated to PHP 8.4 with JIT, benchmarked from 1M to 1B records. Half the hypotheses were wrong: SplFixedArray loses on speed, mmap is 7× slower per call, match equals switch. The other half become survival tools at scale.
- [What 16 Parallel Claude Agents Built Around Themselves: Deconstructing Anthropic's C Compiler Experiment](https://vbcherepanov.com/articles/what-16-parallel-claude-agents-built)
  - Canonical (medium): https://medium.com/@vbcherepanov/what-16-parallel-claude-agents-built-around-themselves-deconstructing-anthropics-c-compiler-f2fa6335b1ca
  - Analysis of Anthropic's experiment running 16 parallel Claude instances to build a C compiler in Rust — what infrastructure had to be invented around the agents because the primitives weren't there yet.
- [The right of an AI agent to stay silent](https://vbcherepanov.com/articles/the-right-of-an-ai-agent-to-stay-silent)
  - Canonical (medium): https://medium.com/@vbcherepanov/the-right-of-an-ai-agent-to-stay-silent-db29c478e577
  - Production AI should optimise for zero confidently-wrong actions, not accuracy percentage. Abstain as a first-class outcome, self-tasking, and a cognitive runtime are the path to trustworthy autonomy.
- [Seven principles of real memory for AI agents](https://vbcherepanov.com/articles/seven-principles-of-real-memory-for-ai-agents)
  - Canonical (medium): https://medium.com/@vbcherepanov/seven-principles-of-real-memory-for-ai-agents-3029d7d877ac
  - Atomic units with lifecycle, strict mode with abstention, causal decision chains, AST-based code identity, internal git, scoring mechanics, and negative memory with rule engines — the recipe behind real cross-session memory.
- [RAG isn't memory. It's Ctrl+F with embeddings.](https://vbcherepanov.com/articles/rag-isnt-memory-its-ctrl-f-with-embeddings)
  - Canonical (medium): https://medium.com/@vbcherepanov/rag-isnt-memory-it-s-ctrl-f-with-embeddings-c461b90ac7b1
  - Vector search is search, not memory. Three failure modes — chunks lose meaning, no structural distinction between facts, and no temporal validity — make RAG hallucinate confidently without knowing what it doesn't know.
- [Why AI-Generated Code Is Technical Debt From Day Zero](https://vbcherepanov.com/articles/why-ai-generated-code-is-technical-debt-from-day-zero)
  - Canonical (medium): https://medium.com/@vbcherepanov/why-ai-generated-code-is-technical-debt-from-day-zero-da3421a73989
  - Patterns that quietly accumulate debt: phantom abstractions, copy-paste-by-prompt, silent error degradation, context amnesia between requests, and decorative tests. Human oversight of architecture and immediate refactoring are the antidote.
- [Your AI Coding Assistant Has Amnesia. Here's How I Fixed It.](https://vbcherepanov.com/articles/your-ai-coding-assistant-has-amnesia)
  - Canonical (medium): https://medium.com/@vbcherepanov/your-ai-coding-assistant-has-amnesia-heres-how-i-fixed-it-a8429f7f7e38
  - Total Agent Memory — an open-source MCP server providing persistent knowledge across sessions. 32 tools spanning storage, self-improvement via error tracking, knowledge graphs, episodic recall, and skill assessment.
- [I Studied the etcd Codebase — and It Changed How I Write PHP](https://vbcherepanov.com/articles/i-studied-the-etcd-codebase)
  - Canonical (medium): https://medium.com/@vbcherepanov/i-studied-the-etcd-codebase-and-it-changed-how-i-write-php
  - Seven architectural principles distilled from etcd's Go codebase, mapped onto everyday PHP/Symfony work: typed contracts, single-responsibility services, composable middleware, observability as architecture, facade APIs, traceable request paths, and explicit dependencies.
- [ClearVibeArchitecture (CVA) — Complete Guide](https://vbcherepanov.com/articles/clear-vibe-architecture)
  - An architectural style for backend systems that combines Hexagonal, Outbox/Inbox, Observability, mandatory Feature Toggles, and flow visualisation.
- [VR Microservices Visualizer](https://vbcherepanov.com/articles/vr-microservices-visualizer)
  - Immersive VR tool for debugging and analysing complex microservice flows with Go, gRPC, and Unity VR — bottlenecks in seconds, request flows in real time.

## Connect

- GitHub: https://github.com/vbcherepanov
- LinkedIn: https://linkedin.com/in/progerinvr
- Medium: https://medium.com/@vbcherepanov
- Dev.to: https://dev.to/vbcherepanov
- YouTube: https://youtube.com/channel/UCVSqFbjRQxNDIqol6cchQ2A

## Optional

- [Full content corpus (Markdown)](https://vbcherepanov.com/llms-full.txt)
- [RSS feed](https://vbcherepanov.com/rss.xml)
- [Sitemap](https://vbcherepanov.com/sitemap.xml)