$ whoami --verbose

const sergei = {
  alias:     "ForeverInLaw",
  role:      "engineer / builder / system plumber",
  org:       "redstone.md — co-founder, open-source R&D (P2P networking, AI moderation)",
  stack:     ["Go", "TypeScript", "Python", "Rust-curious"],
  domains:   ["ai-agents", "e2ee & security", "edge & proxies", "telegram bots", "devtools"],
  shipping:  ["moss", "mosh", "shield", "screenstyler", "whispertocode", "corpcoreai"],
  beliefs:   [
    "code that ships > code that's elegant",
    "private repos > demo repos",
    "if it annoys you twice, automate it",
  ],
  status:    "shipping. always.",
};

▍ what I build

• AI agents & operator tooling — multi-model routing, RAG pipelines (LangChain, FAISS), Telegram-native UX, eval loops
• Security & E2EE — Noise Protocol, MLS/OpenMLS, white-box security audits, authn/authz hardening
• Edge & networking — P2P meshes, NAT traversal, GeoDNS, TCP/UDP reverse proxies, traffic shaping
• Multi-tenant SaaS — Postgres row-level security for tenant isolation, offline-first PWAs, RBAC
• Bots & automation — Discord, Telegram, Steam, Spotify, escrow, support, "do-this-for-me" daemons
• Infra & launchers — config sync, restart APIs, health gateways, deploy pipelines for tiny ARM boxes
• Side-quests — game modding, scrapers, downloaders, anything that scratches an itch in under a weekend

▍ selected work

🌿 moss Embeddable P2P mesh core. Noise XX/IK encrypted transport, topic pubsub with peer scoring, NAT traversal (hole punching, STUN, UPnP, relay fallback). Ships as a C shared library with bindings for C/C++/C#/Python/Rust. ~25k LOC · ~0.86 test-to-source · fuzz + multi-node churn in CI	💬 mosh Desktop E2EE messenger on the moss runtime. OpenMLS (RFC 9420) group encryption, encrypted voice calls (AES-GCM-sealed Opus), live libmoss lifecycle, connection state machine that recovers after failed NAT punches. E2EE text + voice · mosh:// deep links · per-platform bundled runtime
🛡️ shield Self-hosted Telegram moderation platform. Fast-path heuristics (Bayes, similarity, CAS) + slow-path LLM/Vision escalation (OpenAI, Gemini) behind circuit breakers and token budgets. Multi-tenant control plane, live-versioned rule sets, Lua plugins. 240+ commits on top of upstream tg-spam	📸 screenstyler Browser-based screenshot editor & device mockup tool. Two modes: zero-cloud local-first (projects in IndexedDB) or cloud (Neon Postgres, Cloudflare R2, passwordless auth). Undo/redo history, Docker image that self-applies DB migrations. local-first by default · Vitest + Playwright suites
🎙️ whispertocode Cross-platform push-to-talk dictation. Global hotkey capture, NVIDIA Riva Whisper transcription typed straight into any focused input, optional streaming LLM cleanup pass. Typing is slower than thinking out loud. ships as a standalone binary	🤖 CorpCoreAI Telegram bot + WebApp (Next.js) for corporate task flow. NL task intake, LLM-generated titles/subtasks, Postgres via Prisma, role-scoped UI, hourly reminder scheduler with day-of / overdue / manager-escalation alerts. bot + WebApp in one deploy

public list is visible 10%. real firepower lives in private repos — corelogicx, axiomechat, harald, bananalyze, mlefia, owork365, plus a graveyard of profitable internal tools.

▍ stack

languages
backend
frontend
ai / ml
data
infra
daily driver

▍ stats