Georgi Gaydarov — engineering docs hub
Senior network engineering leader building AI-native network operations tooling. Below are the live, single-page docs for my open-source projects — multivendor labs, air-gapped change validation, AI log analysis, an agentic AI pentester, and a 70k-command CLI reference.
Projects
8 open-source projectsargus
Agentic AI penetration tester — reasons, chains, and adapts across network, host, Active Directory and web targets, read-only by default behind a fail-closed 7-layer guardrail. Proof-annotated attack paths, HMAC-audited runs, and Claude / Ollama / offline backends.
Open docs →repo ↗napalm-live-lab
An honest, live multivendor NAPALM coverage matrix — Arista cEOS, Nokia SR Linux & FRR in containerlab — plus a safe-by-default Command Console that runs 2,361 curated commands live against the lab.
Open docs →repo ↗multivendor-ai-network-lab
26-device multivendor AI network lab that closes the loop — detected anomalies auto-remediate, risk-gated (auto / approve / page-out) with RFC 6241 confirmed-commit rollback. Pydantic-AI orchestrator, eval harness with LLM-as-judge, immutable GAIT audit, path trace, intent-drift detector, and CVE scanner.
Open docs →fabric portal ↗multivendor-cli-configurator
70,000+ CLI commands across 17 network vendors & tools — a searchable, deep-linkable cheatsheet with concept-aligned N-vendor compare and per-command NETCONF / ncclient / Ansible drawers. Single HTML, zero deps.
Open docs →live tool ↗aegis
Air-gapped pre-deployment network change validation against a real containerlab digital twin — a self-hosted LLM proposes inside the wall; only a sealed PCI / SOC2 / NIST evidence badge ever leaves. Zero egress.
Open docs →repo ↗netlog-ai
Local, sanitize-first AI network log analyzer (Junos / EOS / FRR) that turns raw syslog and CLI output into prioritized, copy-pastable root-cause playbooks. Runs on a local model or Anthropic Claude.
Open docs →repo ↗claude-mastery-hub
Interactive single-page guide to mastering Claude — the app, Claude Code, the API, MCP, skills, subagents, and hooks. Zero-dependency, deep-linkable reference.
Open docs →repo ↗claude-skill-lint
Zero-dependency linter for Claude Code skills — checks frontmatter, description length, progressive disclosure, dead references, and stale model IDs so a skills library stays lean and loadable.
View repo →